I've found this to be one of the most useful ways to use (at least) GPT-4 for programming. Instead of telling it how an API works, I make it guess, maybe starting with some example code to which a feature needs to be added. Sometimes it comes up with a better approach than I had thought of. Then I change the API so that its code works.
Conversely, I sometimes present it with some existing code and ask it what it does. If it gets it wrong, that's a good sign my API is confusing, and how.
These are ways to harness what neural networks are best at: not providing accurate information but making shit up that is highly plausible, "hallucination". Creativity, not logic.
(The best thing about this is that I don't have to spend my time carefully tracking down the bugs GPT-4 has cunningly concealed in its code, which often takes longer than just writing the code the usual way.)
There are multiple ways that an interface can be bad, and being unintuitive is the only one that this will fix. It could also be inherently inefficient or unreliable, for example, or lack composability. The AI won't help with those. But it can make sure your API is guessable and understandable, and that's very valuable.
Unfortunately, this only works with APIs that aren't already super popular.
> Sometimes it comes up with a better approach than I had thought of.
IMO this has always been the killer use case for AI—from Google Maps to Grammarly.
I discovered Grammarly at the very last phase of writing my book. I accepted maybe 1/3 of its suggestions, which is pretty damn good considering my book had already been edited by me dozens of times AND professionally copy-edited.
But if I'd have accepted all of Grammarly's changes, the book would have been much worse. Grammarly is great for sniffing out extra words and passive voice. But it doesn't get writing for humorous effect, context, deliberate repetition, etc.
The problem is executives want to completely remove humans from the loop, which almost universally leads to disastrous results.
> The problem is executives want to completely remove humans from the loop, which almost universally leads to disastrous results
Thanks for your words of wisdom, which touch on a very important other point I want to raise: often, we (i.e., developers, researchers) construct a technology that would be helpful and "net benign" if deployed as a tool for humans to use, instead of deploying it in order to replace humans. But then along comes a greedy business manager who reckons recklessly that using said technology not as a tool, but in full automation mode, results will be 5% worse, but save 15% of staff costs; and they decide that that is a fantastic trade-off for the company - yet employees may lose and customers may lose.
The big problem is that developers/researchers lose control of what they develop, usually once the project is completed if they ever had control in the first place. What can we do? Perhaps write open source licenses that are less liberal?
The problem here is societal, not technological. An end state where people do less work than they do today but society is more productive is desirable, and we shouldn't be trying to force companies/governments/etc to employ people to do an unnecessary job.
The problem is that people who are laid off often experience significant life disruption. And people who work in a field that is largely or entirely replaced by technology often experience permanent disruption.
However, there's no reason it has to be this way - the fact people having their jobs replace by technology are completely screwed over is a result of the society we have all created together, it's not a rule of nature.
> The problem here is societal, not technological.
I disagree. I think it's both. Yes, we need good frameworks and incentivizes on a economic/political level. But also, saying that it's not a tech problem is the same as saying "guns don't kill people". The truth is, if there was no AI tech developed, we would not need to regulate it so that greed does not take over. Same with guns.
> The truth is, if there was no AI tech developed, we would not need to regulate it so that greed does not take over.
Same could be said for the Internet as we know it too. Literally replace AI with Internet above and it reads equally true. Some would argue (me included some days) we are worse off as a society ~30 years later. That’s also a legitimate case that can be made it was a huge benefit to society too. Will the same be said of AI in 2042?
Oh the web was full of slop long before LLMs arrived. Nothing new. If anything, AI slop is higher quality than was SEO crap. And of course we can't uninvent AI just like we can't unborn a human.
Yes, AI text could be considered higher quality than traditional SEO, but at the same time, it's also very much not, because it always sounds like it might be authoritative, but you could be reading something hallucinated.
In the end, the text was still only ever made to get visitors to websites, not to provide accurate information.
> However, there's no reason it has to be this way - the fact people having their jobs replace by technology are completely screwed over is a result of the society we have all created together, it's not a rule of nature.
I agree. We need a radical change (some version of universal basic income comes to mind) that would allow people to safely change careers if their profession is no longer relevant.
No way that will ever happen when we have a party that thinks Medicare, Medicaid and social security is unnecessary for the poor or middle class. But you better believe all our representatives have that covered for themselves while pretending to serve us (they only serve those that bribe/lobby them)
So it's simple: don't do anything at all about the technology that is the impetus for these horrible disruptions, just completely rebuild our entire society instead.
> the fact people having their jobs replace by technology are completely screwed over is a result of the society we have all created together, it's not a rule of nature.
How did the handloom weavers and spinners handle the rise of the machines?
> How did the handloom weavers and spinners handle the rise of the machines?
In the past, new jobs appeared that the workers could migrate to.
Today, it seems that AI may replace jobs much quicker than before and it's not clear to me which new jobs will be "invented" to balance the loss.
Optimists will say that we have always managed to invent new types of work fast enough to reduce the impact to society, but in my opinion it is unlikely to happen this time. Unless the politicians figure out a way to keep the unemployment content (basic income etc.),
I fear we may end up in a dystopia within our lifetimes. I may be wrong and we could end up in a post scarcity (star trek) world, but if the current ambitions of the top 1% is an indicator, it won't happen unless the politicians create a better tax system to compensate the loss of jobs. I doubt they will give up wealth and influence voluntarily.
I think if we zoom out of the tech and into a bit more of economic the risk I see is that the incumbent hold a lot of advantages and also control the means of production due secondary factors like gpu scarcity.
If we want to draw some parallel this may trigger a robber baron kind of outcome more than an industrial revolution.
The existence of workable open weight models tips me more toward the optimistic outcome
Butthere's trillions at stake now and that must not be discounted it's the kind of wealth accumulation that can easily trigger a war. (And if you thinkit isn't you can look at the oil wars in the 90s and other more recent resources war bring fought in Europe today.
Expect "gpu gap" talks sooner that later, and notice there's a few global power with no horse to race.
> In the past, new jobs appeared that the workers could migrate to.
There was no happy and smooth transition that you seem to allude to. The Luddite movement was in direct response to this: people were dying over this. Factory owners fired or massively reduced wages of workers, replacing many with child workers in precarious and dangerous conditions. In response, the workers smashed the machines that were being used to eliminate their jobs and prevent them from feeding themselves and their families (_not_ the machines that were used to make their jobs easier).
I think you’re describing the principle/agent problem that people have wrestled with forever. Oppenheimer comes to mind.
You make something, but because you don’t own it—others caused and directed the effort—you don’t control it. But the people who control things can’t make things.
Should only the people who can make things decide how they are used though? I think that’s also folly. What about the rest of society affected by those things?
It’s ultimately a societal decision-making problem: who has power, and why, and how does the use of power affect who has power (accountability).
> Grammarly is great for sniffing out extra words and passive voice. But it doesn't get writing for humorous effect, context, deliberate repetition, etc.
> But then along comes a greedy business manager who reckons recklessly
> The big problem is that developers/researchers lose control
if these developers/researchers are being paid by someone else, why should that same someone else be giving up the control that they paid for?
If these developers/researchers are paying the research themselves (e.g., a startup of their own founding), then why would they ever lose control, unless they sell it?
This is a good point. FAANG or whatever you want to call it now has spent billions hovering up a couple generations' best minds who willing sold their intellect to make endless engagement loops.
The problem of those greedy business managers you speak of is that, they don't care how the company does 10 year down the line and I almost feel as if everybody is just doing things which work short term ignoring the long term consequences.
As the comment above said that we need a human in the loop for better results, Well firstly it also depends on human to human.
A senior can be way more productive in the loop than a junior.
So Everybody has just stopped hiring juniors because they cost money and they will deal with the AI almost-slop later/ someone else will deal with it.
Now the current seniors will one day retire but we won't have a new generation of seniors because nobody is giving juniors a chance or that's what I've heard about the job market being brutal.
You're trying to put out a forest fire with an eyedropper.
Stock your underground bunkers with enough food and water for the rest of your life and work hard to persuade the AI that you're not a threat. If possible, upload your consciousness to a starwisp and accelerate it out of the Solar System as close to lightspeed as you can possibly get it.
Those measures might work. (Or they might be impossible, or insufficient.) Changing your license won't.
Alternatively, persuade the AI that you are all-powerful and that it should fear and worship you. Probably a more achievable approach, and there’s precedent for it.
> Alternatively, persuade the AI that you are all-powerful and that it should fear and worship you.
I understand this is a bit deeper into one of the _joke_ threads, but maybe there’s something here?
There is a distinction to be made between artificial intelligence and artificial consciousness. Where AI can be measured, we cannot yet measure consciousness despite that many humans could lay plausible claim to possessing consciousness (being conscious).
If AI is trained to revere or value consciousness while simultaneously being unable to verify it possesses consciousness (is conscious), would AI be in a position to value consciousness in (human) beings who attest to being conscious?
That only works on the AIs that aren't a real threat anyway, and I don't think it helps with the social harm done by greedy business managers with less powerful AIs. In fact, it might worsen it.
Yes, we have the context - our unique lived experience, and are ultimately accountable for our actions. LLMs have no skin. They have no desires, and cannot be punished in any way. No matter how smart they get, we are providing their opportunities to generate value, guidance and iteration, and in the end have to live with the outcomes.
Only some things scale like that. Google's insistence to use the same model everywhere has gained them a deserved reputation as having atrocious support.
Passive voice often adds length, impedes flow, and subtracts the useful info of who is doing something.
Examples:
* Active - concise, complete info: The manager approved the proposal.
* Passive - wordy, awkward: The proposal was approved by the manager.
* Passive - missing info: The proposal was approved. [by who?]
Most experienced writers will use active unless they have a specific reason not to, e.g., to emphasize another element of the sentence, as the third bullet's sentence emphasizes approval.
Many times this is exactly what we want: to emphasize the action instead of who is doing it. It turns out that technical writing is one of the main areas where we want this! So I have always hated this kind of blanket elimination of passive voice.
The passive voice just switches the roles so that the patient is the subject and the agent is the object (e.g. in "The ball was kicked by John," the ball is still the patient despite being the subject). It's just that with English word order, it also switches the places of the things in the sentence.
In languages with more flexible word order, you could just switch the two without passive voice. You could just say the equivalent of "The ball kicked John," with it being clear somehow that the ball is the grammatical object and John the subject, without needing to use the passive voice at all.
Sometimes the missing info is obvious, irrelevant, or intentionally not disclosed, so "The proposal was approved" can be better. Informally we often say, "They approved the proposal," in such cases, or "You approve the proposal" when we're talking about a future or otherwise temporally indefinite possibility, but that's not acceptable in formal registers.
Unfortunately, the resulting correlation between the passive voice and formality does sometimes lead poor writers to use the passive in order to seem more formal, even when it's not the best choice.
E-Prime (short for English-Prime or English Prime, sometimes É or E′) denotes a restricted form of English in which authors avoid all forms of the verb to be.
E-Prime excludes forms such as be, being, been, present tense forms (am, is, are), past tense forms (was, were) along with their negative contractions (isn't, aren't, wasn't, weren't), and nonstandard contractions such as ain't and 'twas. E-Prime also excludes contractions such as I'm, we're, you're, he's, she's, it's, they're, there's, here's, where's, when's, why's, how's, who's, what's, and that's.
Some scholars claim that E-Prime can clarify thinking and strengthen writing, while others doubt its utility.
That's a cool Easter egg page, where the main article text itself is in E-Prime (in use, not in mention), except for where it lists the criticisms and counterarguments - that part has copious amounts of "to be" :)
I've had entire conversations in E-Prime. I found it an interestingly brain-twisting exercise, but still managed to smuggle in all kinds of covert presumptions of equivalence and essential (or analytic) attributes, even though E-Prime's designers intended it to force you to question such things.
Well, I had those conversations a long time ago, but we can describe some general patterns.
We can smuggle in presumptions through the use of attributive adjectives. In the above comment (which you might have noticed I wrote in E-Prime) I mentioned smuggling in "covert presumptions" of "essential attributes". If I had instead written that in assembly language as follows:
I smuggled in presumptions of attributes.
The presumptions were covert.
The attributes were essential.
it would clearly violate E-Prime. And that forces you to ask: does he intend "covert" to represent an essential attribute of those presumptions, or merely a temporary or circumstantial state relative to a particular temporal context? Did he intend "essential" to limit the subjects of discourse to only certain attributes (the essential ones rather than the accidental ones), and within what scope do those attributes have this purported essentiality? Universally, in every possible world, or only within the confines of a particular discourse?
In these particular cases, though, I smuggled in no such presumptions! Both adjectives merely delimit the topic of discourse, to clarify that it does not pertain to overt presumptions or to presumptions of accidental attributes. (As I understand it, Korzybski objects to the "is of predication" not because no predicates exist objectively, but because he doubts the essentiality of any predicates.)
But you can use precisely the same structure to much more nefarious rhetorical ends. Consider, "Chávez kicked the squalid capitalists out of the country." Well, he kicked out all the capitalists! We've smuggled in a covert presumption of essentiality, implying that capitalism entails squalidity. And E-Prime's prohibition on the copula did not protect us at all. If anything, we lose much rhetorical force if we have to explicitly assert their squalidity, using an explicit statement that invites contradiction:
The capitalists are squalid.
We find another weak point at alternative linking verbs. It clearly violates E-Prime to say, "Your mother's face is uglier than a hand grenade," and rightly so, because it projects the speaker's subjective perceptions out onto the world. Korzybski (or Bourland) would prefer that we say, for example, "Your mother's face looks uglier to me than a hand grenade," or possibly, "I see your mother's face as uglier than a hand grenade," thus relativizing the attribute to a single speaker's perception. (He advocated clarity of thought, not civility.)
But we can cheat in a variety of ways that still smuggle in that judgment of essentiality!
Your mother's face turned uglier than a hand grenade.
We can argue this one. Maybe tomorrow, or after her plastic surgery, it will turn pretty again, rather than having ugliness as an essential attribute.
Your mother's face became uglier than a hand grenade.
This goes a little bit further down the line; "became" presupposes a sort of transformation of essence rather than a mere change of state. And English has a variety of verbs that we can use like that. For example, "find", as in "Alsup found Dahmer guilty." Although in that case "find" asserts a state (presumably Dahmer became guilty at some specific time in the past), we can also use it for essential attributes:
I find your mother's face uglier than a hand grenade.
Or lie, more or less, about the agent or speaker:
Your mother's face finds itself uglier than a hand grenade.
And of course we can retreat to attributive adjectives again:
Your mother has a face uglier than a hand grenade.
Your mother comes with an uglier face than a hand grenade.
Or we can simply omit the prepositional phrase from the statement of subjective perception, thus completely erasing the real agent:
Your mother's face looks uglier [...] than a hand grenade.
Korzybski didn't care about the passive voice much, though; E-Prime makes it more difficult but, mostly, not intentionally. As an exception, erasing the agent through the passive voice can misrepresent the speaker's subjective perception as objective:
Your mother's face is found uglier than a hand grenade.
But that still works if we use any of the alternative, E-Prime-permitted passive-voice auxiliary verbs:
Your mother's face gets found uglier than a hand grenade.
As Bourland said, I have "transform[ed] [my] opinions magically into god-like pronouncements on the nature of things".
As another example, notice all the times I've used "as" here. Many of these times smuggle in a covert assertion of essential attributes or even of identity!
But I found it very interesting to notice these things when E-Prime forced me to rethink how I would say them with the copula. It seems like just the kind of mental exercise to heighten my attention to implicit assumptions of identity and essentiality that Korzybski intended.
I wrote the above in E-Prime, by the way. Just for fun.
Yep, just like tritones in music, there is a place for passive voice in writing. But also like tritones, the best general advice is that they should be avoided.
That doesn't make sense. It's like saying that the best general advice about which way to turn when you're driving is to turn right. From your comment at https://news.ycombinator.com/item?id=44493308, and from the fact that you used the passive voice in your comment ("they should be avoided") apparently without noticing, it appears that the reason you have this opinion is that you don't know what the passive voice is in the first place.
I can’t find it, but I remember reading an article a year or two ago with an analysis showing some of the most vocal critics of the passive voice used the passive voice more often than most of their contemporary writers.
If you always turn right at every intersection, you will just go around and around the same block. Which way you should turn depends on where you want to go.
No. That is what roundabouts, curved roads etc are for. Left turns are generally more problematic due to crossing incoming traffic etc.. Hence planning avoids them for good reason and there are much more right turns.
You didn’t originally say anything about always turning right at every intersection and neither did the GP. I had the same two thoughts as GP when I read your analogy.
#2 Is the most pleasant form. The proposal being approved is the most important. #1 Tries to make the manager approving more important then the approval.
That's funny because I read this entirely differently (somewhat dependent on context)
"A decision was made to..." is often code for "The current author didn't agree with [the decision that was made] but it was outside their ability to influence"
Often because they were overruled by a superior, or outvoted by peers.
That's funny, I always thought that meant, "my superior told me I had to do this obviously stupid thing but I'm not going to say my superior was the one who decided this obviously stupid thing." Only occasionally, that is said in a tongue-and-cheek way to refer directly to the speaker as the "superior in charge of the decision."
That reads like several comments I've left in code when I've been told to do something very obviously dumb, but did not want to get tagged with the "why was it done this way?" by the next person reading the code
Usually the passive voice is used at work to emphasize that it was a team/consensus decision, adjacent to the blameless incident management culture. It’s not important that one engineer or PM pushed it, but that ultimately the decision was aligned on and people should be aware.
Although arguably it would be clearer with the active voice and which specific teams / level of leadership aligned on it, usually in the active voice people just use the royal “we” instead for this purpose which doesn’t add any clarity.
Alternatively sometimes I don’t know exactly who made the decision, I just learned it from an old commit summary. So in that case too it’s just important that some people at some time made the decision, hopefully got the right approvals, and here we are.
> Passive - wordy, awkward: The proposal was approved by the manager.
Oh the horror. There are 2 additional words "was" and "by". The weight of those two tiny little words is so so cumbersome I can't believe anyone would ever use those words. WTF??? wordy? awkward?
It's wordy to a high school teacher. Like using "nor" incorrectly it will cause some people's brows to furrow. Always best to be aware of the rules you choose to break.
The problem is that many people have only a poor ability to recognize the passive voice in the first place. This results in the examples being clunky, wordy messes that are bad because they're, well, clunky and wordy, and not because they're passive--indeed, you've often got only a fifty-fifty chance of the example passive voice actually being passive in the first place.
I'll point out that the commenter you're replying to used the passive voice, as did the one they responded to, and I suspect that such uses went unnoticed. Hell, I just rewrote the previous sentence to use the passive voice, and I wonder how many people think recognized that in the first place let alone think it worse for being so written.
> Hell, I just rewrote the previous sentence to use the passive voice
Well, sort of. You used the passive voice, but you didn't use it on any finite verbs, placing your example well outside the scope of the normal "don't use the passive voice" advice.
It would mean that somewhere in your sentence there's a clause headed by a passive verb. A finite verb is one that heads a clause.
This terminology is where we get the name of the "infinitive" form from, by the way.
As a rule of thumb, the nonfinite forms of a verb are its infinitives and participles. jcranmer used a passive participle, but all of his clauses are active. Unnoticed doesn't have a clause around it.
(He might have thought that go unnoticed is a passive form, perhaps of the verb notice (?), in which case that would just be an error.)
Passive can be disastrous when used in contractual situations if the agent who should be responsible for an action isn’t identified. E.g. “X will be done”. I was once burnt by a contract that in some places left it unclear whether the customer or the contractor was responsible for particular tasks. Active voice that identifies the agent is less ambiguous
There was a time when Microsoft Word would treat the passive voice in your writing with the same level of severity as spelling errors or major grammatical mistakes. Drove me absolutely nuts in high school.
Eventually, a feature was added (see what I did there?) that allowed the type of document to be specified, and setting that to ‘scientific paper’ allowed passive voice to be written without being flagged as an error.
Sometimes it's used without thinking, and often the writing is made shorter and clearer when the passive voice is removed. But not always; rewriting my previous sentence to name the agents in each case, as the active voice requires in English, would not improve it. (You could remove "made", though.)
In addition to the points already made, passive voice is painfully boring to read. And it's literally everywhere in technical documentation, unfortunately.
I don't think it's boring. It's easy to come up with examples of the passive voice that aren't boring at all. It's everywhere in the best writing up to the 19th century. You just don't notice it when it's used well unless you're looking for it.
Consider:
> Now we are engaged in a great civil war, testing whether that nation, or any nation so conceived and so dedicated, can long endure.
This would not be improved by rewriting it as something like:
> Now the Confederacy has engaged us in a great civil war, testing whether that nation, or any nation whose founders conceived and dedicated it thus, can long endure.
This is not just longer but also weaker, because what if someone else is so conceiving and so dedicating the nation? The people who are still alive, for example, or the soldiers who just fought and died? The passive voice cleanly covers all these possibilities, rather than just committing the writer to a particular choice of who it is whose conception and dedication matters.
Moreover, and unexpectedly, the passive voice "we are engaged" takes responsibility for the struggle, while the active-voice rephrasing "the Confederacy has engaged us" seeks to evade responsibility, blaming the Rebs. While this might be factually more correct, it is unbefitting of a commander-in-chief attempting to rally popular support for victory.
(Plausibly the active-voice version is easier to understand, though, especially if your English is not very good, so the audience does matter.)
Or, consider this quote from Ecclesiastes:
> For there is no remembrance of the wise more than of the fool for
ever; seeing that which now is in the days to come shall all be forgotten.
You could rewrite it to eliminate the passive voice, but it's much worse:
> For there is no remembrance of the wise more than of the fool for
ever; seeing that everyone shall forget all which now is in the days to come.
This forces you to present the ideas in the wrong order, instead of leaving "forgotten" for the resounding final as in the KJV version. And the explicit agent "everyone" adds nothing to the sentence; it was already obvious.
I think those are important considerations, but it depends even more on what you are attempting to express in the sentence in question. There's plenty of active-voice phrasing in the Gettysburg Address and Ecclesiastes that would not be improved by rewriting it in the passive voice.
>> Now we are engaged in a great civil war, testing whether that nation, or any nation so conceived and so dedicated, can long endure.
> This would not be improved by rewriting it as something like:
>> Now the Confederacy has engaged us in a great civil war [...]
It's technically possible to parse "we are engaged" as a verb in the passive voice.
But it's an error to think that's how you should parse it. That clause is using the active verb be, not the passive verb engage; it's fully parallel to "Now we are happy".
Thats not passive voice. Passive voice is painfully boring to read is active. The preamble can be read like “however”, and is unnecessary; what a former editor of mine called “throat-clearing words”.
"the points already made" is what is known as the "bare passive", and yes, it is the passive voice. You can see e.g. https://languagelog.ldc.upenn.edu/nll/?p=2922 for a more thorough description of the passive voice.
As I said elsewhere, one of the problems with the passive voice is that people are so bad at spotting it that they can at best only recognize it in its worst form, and assume that the forms that are less horrible somehow can't be the passive voice.
I'm not sure this is a "bare passive" like the beginning of "The day's work [being] done, they made their way back to the farmhouse," one of the bare-passive examples at your link. An analogous construction would be, "The points already [being] made, I ceased harassing the ignorant". But in "In addition to the points already made" this case "the point already made" is not a clause; it's a noun phrase, the object of the preposition "to". Its head is "points", and I believe that "made" is modifying that head.
Can you insert an elided copula into it without changing the meaning and grammatical structure? I'm not sure. I don't think so. I think "In addition to the points already being made" means something different: the object of the preposition "to" is now "being", and we are going to discuss things in addition to that state of affairs, perhaps other things that have happened to the points (being sharpened, perhaps, or being discarded), not things in addition to the points.
I agree that that has the same meaning, but I think it may have a different grammatical structure, with an entire subordinate clause that was missing from the original. Since the voice of a verb is a grammatical rather than semantic question, this seems relevant; "in addition to the points people have already made" is also (probably) semantically equivalent but unquestionably uses the active voice.
Yes, the verb "is" in "Passive voice is painfully boring to read" is in the active voice, not the passive voice. But umanwizard was not saying that "is" was in the passive voice. Rather, they were saying that the past participle "made", in the phrase "the points already made", is a passive-voice use of the verb "make".
I don't know enough about English grammar to know whether this is correct, but it's not the assertion you took issue with.
Why am I not sure it's correct? If I say, "In addition to the blood so red," I am quite sure that "red" is not in the passive voice, because it's not even a verb. It's an adjective. Past participles are commonly used as adjectives in English in contexts that are unambiguously not passive-voice verbs; for example, in "Vito is a made man now," the past participle "made" is being used as an attributive adjective. And this is structurally different from the attributive-verb examples of "truly verbal adjectives" in https://en.wikipedia.org/wiki/Attributive_verb#English, such as "The cat sitting on the fence is mine," and "The actor given the prize is not my favorite;" we could grammatically say "Vito is a man made whole now". That page calls the "made man" use of participles "deverbal adjectives", a term I don't think I've ever heard before:
> Deverbal adjectives often have the same form as (and similar meaning to) the participles, but behave grammatically purely as adjectives — they do not take objects, for example, as a verb might. For example: (...) Interested parties should apply to the office.
So, is "made" in "the points already made" really in passive voice as it would be in "the points that are already made", is it deverbal as it would be in "the already-made points" despite its positioning after the noun (occasionally valid for adjectives, as in "the blood so red"), or is it something else? I don't know. The smoothness of the transition to "the points already made by those numbskulls" (clearly passive voice) suggests that it is a passive-voice verb, but I'm not sure.
> But umanwizard was not saying that "is" was in the passive voice. Rather, they were saying that the past participle "made", in the phrase "the points already made", is a passive-voice use of the verb "make".
> I don't know enough about English grammar to know whether this is correct, but it's not the assertion you took issue with.
The most natural interpretation is indeed that the participle made is being used as a full participle and not as a zero-derived adjective. For example, you could give it a really strong verbal sense by saying "the points already made at length [...]" or "the points made so many times [...]".
> So, is "made" in "the points already made" really in passive voice as it would be in "the points that are already made"
Though I wouldn't say the same thing there; if you say "the points that are already made", that pretty much has to be an adjective. If you want it to be a passive verb, go with "the points that have already been made".
Anyway, I would be really surprised if die-hard thoughtless style prescriptivists thought that the advice "don't use the passive voice" was meant to apply to participles. It's a quibble that you don't care about and they don't care about or understand. You're never going to get anywhere with someone by telling them they mean something they know perfectly well they don't mean.
> Anyway, I would be really surprised if die-hard thoughtless style prescriptivists thought that the advice "don't use the passive voice" was meant to apply to participles.
> The relevance of participles is that a passive clause always has its verb in a participial form.
So, what are you saying they do think it was meant to apply to, if every passive clause always includes a participle? I'm confused.
With respect to:
> Though I wouldn't say the same thing there; if you say "the points that are already made", that pretty much has to be an adjective. If you want it to be a passive verb, go with "the points that have already been made".
the passive-clause examples given in Pullum's blog post I linked above include "Each graduate student is given a laptop," which sounds structurally identical to your example (except that an indirect object is present, showing that it cannot be an adjective) and clarifies:
> The verb was doesn't really add any meaning, but it enables the whole thing to be put into the preterite tense so that the event can be asserted to have occurred in the past. Changing was to is would put the clause into the present tense, and replacing it by will be or is going to be would permit reference to future time; but the passive VP damaged by storms would stay the same in each case. (Notice, the participle damaged does not itself make any past time reference, despite the name "past participle".)
So it sounds like your grammatical analysis is explicitly contradicting Pullum's, which probably means you're wrong, but I'm not sure I understand it.
>> The relevance of participles is that a passive clause always has its verb in a participial form.
> So, what are you saying they do think it was meant to apply to, if every passive clause always includes a participle? I'm confused.
OK, you're confused.
In the general case, an English verb has five forms†: "plain form" [go], "preterite form" [went], "present third-person singular form" [goes], "-ing form" [going], and "-en form" [gone].
The last two of those are participial forms.
It is true that a passive clause always has its verb in a participial form. We can be even more specific than that: the verb is always in -en form. This is true without exception because passive markers occur last in the sequence of auxiliary verbs that might modify a primary verb, and therefore always directly control the form of the primary verb.
It is not true that a passive clause always includes a participle, except in the sense of the name we give to the form of the verb. -ing and -en are "participial forms" because the verb takes one of those forms when it is a participle. But it can also take them for other reasons.
> the passive-clause examples given in Pullum's blog post I linked above include "Each graduate student is given a laptop," which sounds structurally identical to your example
Sure. If you wanted to put the present passive third-person plural form of make in that sentence, that form† would be are made. The sentence would have all the same words in the same order.
But that would make no semantic sense. For a point to be "already made", as opposed to having "already" been "made", you need to interpret made as an adjective, describing the state in which the point currently exists. The temporal structure of "each graduate student is given a laptop" differs from that of "in addition to the points that are already made" in a way that allows the present noncontinuous form of the verb. I don't think that works for "the points that are already made"; if I try to interpret that as a passive verb in the present tense, I get a strong sense that the sentence is malformed.
† You might notice that these two uses of the word form conflict with each other. The fact that form is used in both of these ways is why I'm annoyed at your comment conflating "participle" with "participial form". "Participle" is appropriate when you're talking about inflecting a verb according to how you want to use it in a sentence; it is a concern with the language's grammar. "Participial form" is appropriate when you're talking about the actual tokens that can appear in a sentence, with no regard to what they might mean or how they might be used; it is a concern with what you might think of as the language's "anatomy".
To be fair, current scientific papers are full of utterly terrible writing. If you read scientific papers from a century and a half ago, a century ago, half a century ago, and today, you'll see a continuous and disastrous decline in readability, and I think some of that is driven by pressure to strictly follow genre writing conventions. One of those conventions is using the passive voice even when the active voice would be better.
I suspect that the disastrous results being envisioned are somewhat more severe than not being able to tell who wrote which memo. I understood the author to be suggesting things more like bankruptcy, global warfare, and extermination camps. But it's admittedly ambiguous.
Criticisms are almost always read by the reader as criticisms of the OP's actions. If you're agreeing with somebody as you appear to be here, you should probably make that more explicit.
I used this to great success just this morning. I told the AI to write me some unit tests. It flailed and failed badly at that task. But how it failed was instructive, and uncovered a bug in the code I wanted to test.
It used nonsensical parameters to the API in way that I didn't realize was possible (though obvious in hindsight). The AI got confused; it didn't think the parameters were nonsensical. It also didn't quite use them in the way that triggered the error. However it was close enough for me to realize that "hey, I never though of that possibility". I needed to fix the function to return a proper error response for the nonsense.
It also taught me to be more careful about checkpointing my work in git before letting an agent go wild on my codebase. It left a mess trying to fix its problems.
Many many python image-processing libraries have an `imread()` function. I didn't know about this when designing our own bespoke image-lib at work, and went with an esoteric `image_get()` that I never bothered to refactor.
When I ask ChatGPT for help writing one-off scripts using the internal library I often forget to give it more context than just `import mylib` at the top, and it almost always defaults to `mylib.imread()`.
I don't know if there's an earlier source, but I'm guessing Matlab originally popularized the `imread` name, and that OpenCV (along with its python wrapper) took it from there, same for scipy. Scikit-image then followed along, presumably.
As someone not familiar with these libraries, image_get or image_read seems much clearer to me than imread. I'm wondering if the convention is worse than your instinct in this case. Maybe these AI tools will push us towards conventions that aren't always the best design.
image_get is clearer—unless you've used Matlab, Octave, matplotlib, SciPy, OpenCV, scikit-learn, or other things that have copied Matlab's interface. In that case, using the established name is clearer.
(Unless, on the gripping hand, your image_get function is subtly different from Matlab's imread, for example by not returning an array, in which case a different name might be better.)
That's a perfect example! I wonder if changing it would be an improvement? If you can just replace image_get with imread in all the callers, maybe it would save your team mental effort and/or onboarding time in the future.
I strongly prefer `image_get/image_read` for clarity, but I would just stump in a method called `imread` which is functionally the same and hide it from the documentation.
In a similar vein, some of my colleagues have been feeding their scientific paper methods sections to LLMs and asking them to implement the method in code, using the LLM's degree of success/failure as a vague indicator of the clarity of the method description.
This is similar to an old HCI design technique called Wizard of Oz by the way, where a human operator pretends to be the app that doesn’t exist yet. It’s great for discovering new features.
> Sometimes it comes up with a better approach than I had thought of. Then I change the API so that its code works.
“Sometimes” being a very important qualifier to that statement.
Claude 4 naturally doesn’t write code with any kind of long term maintenance in-mind, especially if it’s trying to make things look like what the less experienced developers wrote in the same repo.
Please don’t assume just because it looks smart that it is. That will bite you hard.
Even with well-intentional rules, terrible things happen. It took me weeks to see some of it.
In essence, a LLM is a crystallisation of a large corpus human opinion and you are using that to focus group your API as it is representative of a reasonable third party perspective?
Yeah, basically. For example, it's really good at generating critical HN comments. Whenever I have a design or an idea I formulate it to GPT and ask it to generate a bunch of critical HN comments. It usually points out stuff I hadn't considered, or at least prepares me to think about and answer the tough questions.
> and being unintuitive is the only one that this will fix
That's also how I'm approaching it. If all the condensed common wisdom poured into the model's parameters says that this is how my API is supposed to work to be intuitive, how on earth do I think it should work differently? There needs to be a good reason (like composability, for example). I break expectations otherwise.
I've played with a similar idea for writing technical papers. I'll give an LLM my draft and ask it to explain back to me what a section means, or otherwise quiz it about things in the draft.
I've found that LLMs can be kind of dumb about understanding things, and are particularly bad at reading between the lines for anything subtle. In this aspect, I find they make good proxies for inattentive anonymous reviewers, and so will try to revise my text until even the LLM can grasp the key points that I'm trying to make.
In both cases, you might get extra bonus usability if the reviewers or the API users actually give your output to the same LLM you used to improve the draft. Or maybe a more harshly quantized version of the same model, so it makes more mistakes.
This was a big problem starting out writing MCP servers for me.
Having an LLM demo your tool, then taking what it does wrong or uses incorrectly and adjusting the API works very very well. Updating the docs to instruct the LLM on how to use your tool does not work well.
> I don't have to spend my time carefully tracking down the bugs GPT-4 has cunningly concealed in its code
If anyone is stuck in this situation, give me a holler. My Gmail username is the same as my HN username. I've always been the one to hunt down my coworkers' bugs, and I think I'm the only person on the planet will finds it enjoyable to find ChatGPT'S oversights and sometimes seemingly malicious intent.
I'll charge you, don't get me wrong, but I'll save you time, money, and frustration. And future bug reports and security issues.
> These are ways to harness what neural networks are best at: not providing accurate information but making shit up that is highly plausible, "hallucination". Creativity, not logic.
This is also similar to which areas TD-Gammon excelled at in Backgammon.
Which is all pretty amusing, if you compare it to how people usually tended to characterise computers and AI, especially in fiction.
That's closer to simply observing the mean. For an analogy, it's like waiting to pave a path until people tread the grass in a specific pattern. (Some courtyard designers used to do just that. Wait to see where people were walking first.)
Making things easy for Chat GPT means making things close to ordinary, average, or mainstream. Not creative, but can still be valuable.
Great point. Also, it may not be the best possible API designer in the world, but it sure sounds like a good way to forecast what an _average_ developer would expect this API to look like.
Often I've started with some example code that invokes part of the API, but not all of it. Or in C I can give it the .h file, maybe without comments.
Sometimes I can just say, "How do I use the <made-up name> API in Python to do <task>?" Unfortunately the safeguards against hallucinations in more recent models can make this more difficult, because it's more likely to tell me it's never heard of it. You can usually coax it into suspension of disbelief, but I think the results aren't as good.
I think you’re missing the OP’s point. They weren’t saying that the goal is to modify their APIs just to appease an LLM. It’s that they ask LLMs to guess what the API is and use that as part of their design process.
If you automatically assume that what the LLM spits out is what the API ought to be then I agree that that’s bad engineering. But if you’re using it to brainstorm what an intuitive interface would look like, that seems pretty reasonable.
Yes, that's a bonus. In fact, I've found it worthwhile to prompt it a few times to get several different guesses at how things are supposed to work. The super lazy way is to just say, "No, that's wrong," if necessary adding, "Frotzl2000 doesn't have an enqueueCallback function or even a queue."
Of course when it suggests a bad interface you shouldn't implement it.
From my perspective that’s fascinatingly upside down thinking that leads to you asking to lose your job.
AI is going to get the hang of coding to fill in the spaces (i.e. the part you’re doing) long before it’s able to intelligently design an API. Correct API design requires a lot of contextual information and forward planning for things that don’t exist today.
Right now it’s throwing spaghetti at the wall and you’re drawing around it.
I find it's often way better than API design than I expect. It's seen so many examples of existing APIs in its training data that it tends to have surprisingly good "judgement" when it comes to designing a new one.
Even if your API is for something that's never been done before, it can usually still take advantage of its training data to suggest a sensible shape once you describe the new nouns and verbs to it.
Maybe. So far it seems to be a lot better at creative idea generation than at writing correct code, though apparently these "agentic" modes can often get close enough after enough iteration. (I haven't tried things like Cursor yet.)
I agree that it's also not currently capable of judging those creative ideas, so I have to do that.
This sort of discourse really grinds my gears. The framing of it, the conceptualization.
It's not creative at all, any more than taking the sum of text on a topic, and throwing a dart at it. It's a mild, short step beyond a weighted random, and certainly not capable of any real creativity.
Myriads of HN enthusiasts often chime in here "Are humans any more creative" and other blather. Well, that's a whataboutism, and doesn't detract from the fact that creative does not exist in the AI sphere.
I agree that you have to judge its output.
Also, sorry for hanging my comment here. Might seem over the top, but anytime I see 'creative' and 'AI', I have all sorts of dark thoughts. Dark, brooding thoughts with a sense of deep foreboding.
Point taken but if slushing up half of human knowledge and picking something to fit into the current context isn't creative then humans are rarely creative either.
> Well, that's a whataboutism, and doesn't detract from the fact that creative does not exist in the AI sphere.
Pointing out that your working definition excludes reality isn't whataboutism, it's pointing out an isolated demand for rigor.
If you cannot clearly articulate how human creativity (the only other type of creativity that exists) is not impugned by the definition you're using as evidence that creativity "does not exist in the AI sphere", you're not arguing from a place of knowledge. Your assertion is just as much sophistry as the people who assert it is creativity. Unlike them, however, you're having to argue against instances where it does appear creative.
For my own two cents, I don't claim to fully understand how human creativity emerges, but I am confident that all human creative works rest heavily on a foundation of the synthesis of author's previous experiences, both personal and of others' creative works - and often more heavily the latter. If your justification for a lack of creativity is that LLMs are merely synthesizing from previous works, then your argument falls flat.
"Whataboutism" is generally used to describe a more specific way of pointing out an isolated demand for rigor—specifically, answering an accusation of immoral misconduct with an accusation that the accuser is guilty of similar immoral misconduct. More broadly, "whataboutism" is a term for demands that morality be judged justly, by objective standards that apply equally to everyone, rather than by especially rigorous standards for a certain person or group. As with epistemic rigor, the great difficulty with inconsistent standards is that we can easily fall into the trap of applying unachievable standards to someone or some idea that we don't like.
So it makes some sense to use the term "whataboutism" for pointing out an isolated demand for rigor in the epistemic space. It's a correct identification of the same self-serving cognitive bias that "whataboutism" targets in the space of ethical reasoning, just in a different sphere.
There's the rhetorical problem that "whataboutism" is a derogatory term for demanding that everyone be judged by the same standards. Ultimately that makes it unpersuasive and even counterproductive, much like attacking someone with a racial slur—even if factually accurate, as long as the audience isn't racist, the racial slur serves only to tar the speaker with the taint of racism, rather than prejudicing the audience against its nominal target.
In this specific case, if you concede that humans are no more creative than AIs, then it logically follows that either AIs are creative to some degree, or humans are not creative at all. To maintain the second, you must adopt a definition of "creativity" demanding enough to exclude all human activity, which is not in keeping with any established use of the term; you're using a private definition, greatly limiting the usefulness of your reasoning to others.
And that is true even if the consequences of AIs being creative would be appalling.
I'll play with your tact in this argument, although I certain do not agree it is accurate.
You're asserting that creativity is a meld of past experience, both personal and the creative output of others. Yet this really doesn't jive, as an LLM does not "experience" anything. I would argue that raw knowledge is not "experience" at all.
We might compare this to the university graduate, head full of books and data jammed therein, and yet that exceptionally well versed graduate needs "experience" in a job for quite some time, before having any use.
The same may be true of learning how to do anything, from driving, to riding a bike, or just being in conversations with others. Being told, on paper (or as part of your baked in, derived "knowledge store") things, means absolutely nothing in terms of actually experiencing them.
Heck, just try to explain sex to someone before they've experienced it. No matter the literature, play, movie or act performed in front of them, experience is entirely different.
And an AI does not experience the universe, nor is it driven by the myriad of human totality, from the mind o'lizard, to the flora/fauna in one's gut. There is no motive driving it, for example it does not strive to mate... something that drives all aspect of mammalian behaviour.
So intertwined with the mating urge is human experience, that it is often said that all creativity derives from it. The sparrow dances, the worm wiggles, and the human scores 4 touchdowns in one game, thank you Al.
Comparatively, an LLM does not reason, nor consider, nor ponder. It is "born" with full access to all of its memory store, has data spewed at it, searches, responds, and then dies. It is not capable of learning in any stream of consciousness. It does not have memory from one birth to the next, unless you feed its own output back at it. It can gain no knowledge, except from "context" assigned at birth.
An LLM, essentially, understands nothing. It is not "considering" a reply. It's all math, top to bottom, all probability, taking all the raw info it has an just spewing what fits next best.
Experiences are not materially different from knowledge once they are both encoded as memories. They're both just encoded in neurons as weights in their network of connections. But let's assume there is some ineffable difference between firsthand and secondhand experience, which fundamentally distinguishes the two in the brain in the present.
The core question here, then, is why you are so certain that "creativity" requires "experience" beyond knowledge, and why knowledge is insufficient? What insight do you have into the human mind that top neuroscientists lack that grants you this gnosticism on how creativity definitely does and does not work?
Because, if you'll permit me to be crude, some of the best smut I've read has been by people I'm certain have never experienced the act. Their writing has been based solely on the writings of others. And yet, knowledge alone is more than enough for them to produce evocative creative works.
And, to really hammer in a point (please forgive the insulting tone):
>It's all math, top to bottom, all probability, taking all the raw info it has an just spewing what fits next best.
You are just biology, top to bottom, just electrical signals, taking all the raw info your nerves get, matching patterns and just spewing what fits next best.
Calling LLMs "just math" -- that's not creative, it's part of your input that you're predicting fits the likely next argument.
You didn't "reason, consider, or ponder" whether I would find that argument convincing or be able to immediately dismiss it because it holds no weight.
You're simply being a stochastic parrot, repeating the phrases you've heard.
...Etcetera. Again, apologies for the insult. But the point I am continually trying to make is that all of the arguments everyone tries to make about it not reasoning, not thinking, not having creativity -- they all are things that can and do apply to almost every human person, even intelligent and articulate ones like you or I.
When it comes down to it, your fundamental argument is that you do not believe that a machine can possibly have the exceptional qualities of the human mind, for some ineffable reason. It's all reasoning backwards from there. Human creativity must require human-like experience, the ability to grow, and a growing context cannot possibly suffice, because you've already decided on your conclusion.
(Because, perhaps, it would be too unsettling to admit that the alien facsimile of intelligence that we've created might have actual intelligence -- so you refuse that possibility)
Experiences are not materially different from knowledge once they are both encoded as memories.
The storage medium is not relevant here, and actual experience has sensorially laden additions which are not currently possible from reading. Memories are laid down differently as well.
In terms of what top neuroscientists know or do not know, you pulled such out of your back pocket, and threw them into onto the floor, perhaps you should describe precisely how they negate what I am saying?
Is there consensus? LLM is indeed creative?
What you seem to be missing here, is I am not railing against machine intelligence, nor creativity. It is merely that an LLM is not it, and will never become it. This is no different than an argument over whether to use sysvinit or systemd, it is a discussion of technical capabilities of a technology.
LLMs may become a backing store, a "library" of sorts for any future AGI to use as data, knowledge, an exceptionally effective method to provide a wikipedia-ish like, non-sql backed data source.
But they provide no means for cognition.
And creativity requires cognition. Creativity is a conscious process, for it requires imagination, which is an offshoot of a conscious process. Redefining "creativity" to exclude the conscious process negates its very meaning.
You can say "Wow, this appears to be creative", and it may appear to be creative, yet without cognition the act is simply not possible. None would dare say that a large Goldberg machine, which spits out random answers dependent upon air currents was producing creative ideas.
Some may say "What a creative creation this machine is!", but none would attribute creativity to the output of any algorithmic production by that machine, and this is what we have here.
Should we derive a method of actual conscious cognition in a mind not flesh, so be it. Creativity may occur. But as things stand now, a mouse provides more creativity than an LLM, the technology is simply not providing the underlying requirements. There is no process for consciousness.
There are ways to provide for this, and I have pondered them (again, I'm validating here that it's not "oh no, machines are NOT thinking!!", but instead "LLMs aren't that").
One exceptionally rough and barely back-of-napkin concept, would be sleep.
No I am not trying to mimic the human mind here, but when the concept is examined front to end, the caboose seems to be 'sleep'. Right now, the problem is how we bring each LLM onto a problem. We simply throw massive context at it, and then allow it to proceed.
Instead, we need to have a better context window. Maybe we should call this 'short term' memory. An LLM is booted, and responds to questions, but has a floating context window which never shrinks. Its "context window" is not cleared. Perhaps we use a symbolic database, or just normal SQL with fluff modz, but we allow this ongoing context window to exist and grow.
After a point, this short term memory will grow too large to actively swap in/out of memory. Primarily, even RAM has bandwidth limits and is a detractor on response speed and energy requirements per query.
So -- the LLM "goes to sleep". During that time, the backend is converted to LLM, or I suppose in this case a small language model. We now have a fuzzy, RRD almost conversion of short-term memory to long-term memory, yet one which enables some very important things.
That being, an actual capacity to learn from interaction.
The next step is to expand that capability, and the capabilities of an LLM with senses. I frankly think the best here is real, non-emulated robotic control. Give the LLM something to manipulate, as well as senses.
At that point, we should inject agency. A reason to exist. With current life, the sole primary reason is "reproduce". Everything else has derived from that premise. I spoke of the mating urge, we should recreate this here.
(Note how non-creative this all actually is, yet it seems valid to me... we're just trying to provide what we know works as a starting base. It does not mean that we cannot expand the creation of conscious minds into other methods, once we have a better understanding and success.)
There are several other steps here, which are essential. The mind must, for example, reload its "long-term memory" backing store SLM, and when "sleep" comes, overlay new short-term thoughts over long-term. This is another fuzzy process, and it would be best to think of it(though technically not accurate) as unpacking its SLM, overlaying new thoughts, and creating an entirely new SLM. As its short-term memory would have output derived from the LLM, plus overlaid SLM, its short-term memory would be providing output derived from its prior SLM.
So there is a form of continuity here.
So we have:
* A mind which can retain information, and is not merely popping into creation with no stored knowledge, and killed at each session end
* That same mind has a longer term memory, which allows for ongoing concept modification and integration, eg new knowledge affecting the "perception" of old knowledge
* That SLM will be "overlaid" on top of its LLM, meaning experiences derived during waking moments will provide more context to an LLM (that moment when you ride a bike, and comprehend how all the literature you read, isn't the same as doing? That moment where you link the two? That's in the SLM, and the SLM has higher priority)
* A body (simple as it may be), which allows access to the environment
* Senses fed into the process
* Agency, as in, "perform to mate", with aspects of "perform" being "examine, discover, be impressive" that sort of thing
I think, this overlaid SLM, along with actual empirical data would provide a more apt method to simulate some form of consciousness. It would at least allow a stream of consciousness, regardless of whatever debates we might have about humans dying when they sleep (which makes no sense, as the brain is incredibly active during sleep, and constantly monitoring the surroundings for danger).
I'd speak more along the sensory aspects of this, but it's actually what I'm working on right now.
But what's key here, is independent data and sensory acquisition. I see this is a best-able way to kickstart a mind.
> Hallucinations can sometimes serve the same role as TDD. If an LLM hallucinates a method that doesn’t exist, sometimes that’s because it makes sense to have a method like that and you should implement it.
>> Hallucinations can sometimes serve the same role as TDD. If an LLM hallucinates a method that doesn’t exist, sometimes that’s because it makes sense to have a method like that and you should implement it.
A detailed counterargument to this position can be found here[0]. In short, what is colloquially described as "LLM hallucinations" do not serve any plausible role in software design other than to introduce an opportunity for software engineers to stop and think about the problem being solved.
Did you mean to post a different link? The article you linked isn’t a detailed counterargument to my position and your summary of it does not match its contents either.
I also don’t see the relevance of Clarke’s third law.
Beware, the feature in OP isn't something that people would have found useful, it's not like chatgpt assigned to OP's business a request from a user in some latent consumer-provider space, as if chatgpt were some kind of market maker connecting consumers with products, like a google with organic content or ads, or linkedin or producthunt.
No, what actually happened is that OP developed a type of chatgpt integration, and a shitty one at that, chatgpt could have just directed the user to the site and told them to upload that image to OP's site. But it felt it needed to do something with the image, so it did.
There's no new value add here, at least yet, maybe if users started requesting changes to the sheet I guess, not what's going on.
> the feature in OP isn't something that people would have found useful
This doesn’t seem likely. The utility is pretty obvious.
> chatgpt could have just directed the user to the site and told them to upload that image to OP's site.
What image? Did you think the first image shown is what is being entered into ChatGPT? It’s not. That’s what the site expects to be uploaded to them. There’s no indication that the ChatGPT users are scanning tabs. ChatGPT is producing ASCII tabs, but we aren’t shown what input it is in response to.
The music notation tool space is balkanized in a variety of ways. One of the key splits is between standard music notation and tablature, which is used for guitar and a few other instruments. People are generally on one side or another, and the notation is not even fully compatible - tablature covers information that standard notation doesn't, and vice versa. This covers fingering, articulations, "step on fuzz pedal now," that sort of thing.
The users are different, the music that is notated is different, and for the most part if you are on one side, you don't feel the need to cross over. Multiple efforts have been made (MusicXML, etc.) to unify these two worlds into a superset of information. But the camps are still different.
So what ChatGPT did is actually very interesting. It hallucinated a world in which tab readers would want to use Soundslice. But, largely, my guess is they probably don't....today. In a future world, they might? Especially if Soundslice then enables additional features that make tab readers get more out of the result.
I don't fully understand your comment, but Soundslice has had first-class support for tablature for more than 10 years now. There's an excellent built-in tab editor, plus importers for various formats. It's just the ASCII tab support that's new.
I wonder if LLMs will stimulate ASCII formats for more things, and whether we should design software in general to be more textual in order to work better with LLMs.
I’m not super familiar with Soundslice. But all the tab users I know use guitar pro or maybe ultimate guitar, and none of them can read standard notation on its own. Does Soundslice have a lot of tab-first users?
Yes, Soundslice has a ton of tab-first users. And in fact the primary reason I founded the site was to scratch my own itch of being able to create tab that's synced with real audio recordings. (I'm a guitarist myself.)
When I read the blog post, I thought it was saying that Soundslice didn't have any tab support.
The comparison of "we expect this (classical notation screenshot) but instead got this (ascii tab screenshot)" made me think that the only thing Soundslice supported was classical notation.
Definitely a subtle distinction there. Soundslice supports tab in many formats (MusicXML, Guitar Pro, PowerTab, TuxGuitar, PDF/images of published tab, or tab notated directly in the Soundslice editor) but didn't support ASCII format yet.
I think folks have taken the wrong lesson from this.
It’s not that they added a new feature because there was demand.
They added a new feature because technology hallucinated a feature that didn’t exist.
The savior of tech, generative AI, was telling folks a feature existed that didn’t exist.
That’s what the headline is, and in a sane world the folks that run ChatGPT would be falling over themselves to be sure it didn’t happen again, because next time it might not be so benign as it was this time.
> in a sane world the folks that run ChatGPT would be falling over themselves to be sure it didn’t happen again
This would be a world without generative AI available to the public, at the moment. Requiring perfection would either mean guardrails that would make it useless for most cases, or no LLM access until AGI exists, which are both completely irrational, since many people are finding practical value in its current imperfect state.
The current state of LLM is useful for what it's useful for, warnings of hallucinations are present on every official public interface, and its limitations are quickly understood with any real use.
Nearly everyone in AI research is working on this problem, directly or indirectly.
> If “don’t hallucinate” is too much to ask then ethics flew out the window long ago.
Those sentences aren't compatible.
> but hallucination is a major issue
Again, every official public AI interface has warnings/disclaimers for this issue. It's well known. It's not some secret. Every AI researcher is directly or indirectly working on this.
> is in the opposite direction of the “goal” of AGI
This isn't a logical statement, so it's difficult to respond to. Hallucination isn't a direction that's being headed towards, it's being actively, with intent and $$$, headed away from.
My web browser isn't perfect, but it does not hallucinate inexistent webpages. It sometimes crashes, it sometimes renders wrong, it has bugs and errors. It does not invent plausible-looking information.
There really is a lot middle gound between perfect and "accept anything we give you, no matter how huge the problems".
> It does not invent plausible-looking information.
This is where your analogy is falling apart; of course web browsers do not "invent plausible-looking information" because they don't invent anything in the first place! Web browsers represent a distinct set of capabilities, and as you correctly pointed out, these are often riddled with bugs and errors. If I was making a browser analogy, I would point towards fingerprinting; most browsers reveal too much information about any given user and system, either via cross-site cookies, GPU prints, and whatnot. This is an actual example where "ethics flew out the window long ago."
As the adjacent commenter pointed out: different software, different failure modes.
> Requiring perfection would either mean guardrails that would make it useless for most cases, or no LLM access until AGI exists
What?? What does AGI have to do with this? (If this was some kind of hyperbolic joke, sorry, i didn't get it.)
But, more importantly, the GP only said that in a sane world, the ChatGPT creators should be the ones trying to fix this mistake on ChatGPT. After all, it's obviously a mistake on ChatGPT's part, right?
That was the main point of the GP post. It was not about "requiring perfection" or something like that. So please let's not attack a straw man.
Their requirement is no hallucinations [1], also stated as "be sure it didn't happen again" in the original comment. If you define a hallucination as something that wasn't in the training data, directly or indirectly (indirectly being something like an "obvious" abstract concept), then you've placed a profound constraint on the system, requiring determinism. That requirement fundamentally, by the non-deterministic statistics that these run on, means you cannot use an LLM, as they exist today. They're not "truth" machines - use a database instead.
Saying "I don't know", with determinism is only slightly different than saying "I know" with determinism, since it requires being fully aware of what you do know, not at a fact level, but at a conceptual/abstract level. Once you have a system that fully reasons about concepts, is self aware of its own knowledge, and can find the fundamental "truth" to answer a question with determinism, you have something indistinguishable from AGI.
Of course, there's a terrible hell that lives between those two, in the form of: "Error: Question outside of known questions." I think a better alternative to this hell would be a breakthrough that allowed "confidence" to be quantified. So, accept that hallucinations will exist, but present uncertainty to the user.
You sound like all the naysayers when Wikipedia was new. Did you know anybody can go onto Wikipedia and edit a page to add a lie‽ How can you possibly trust what you read on there‽ Do you think Wikipedia should issue groveling apologies every time it happens?
Meanwhile, sensible people have concluded that, even though it isn’t perfect, Wikipedia is still very, very useful – despite the possibility of being misled occasionally.
> despite the possibility of being misled occasionally.
There is a chasm of difference between being misled occasionally (Wikipedia) and frequently (LLMs). I don’t think you understand how much effort goes on behind the scenes at Wikipedia. No, not everyone can edit every Wikipedia page willy-nilly. Pages for major political figures often can only be edited with an account. IPs like those of iCloud Private Relay are banned and can’t anonymously edit the most basic of pages.
Furthermore, Wikipedia was always honest about what it is from the start. They managed expectations, underpromised and overdelivered. The bozos releasing LLMs talk about them as if they created the embryo of god, and giving money to their religion will solve all your problems.
A technology can be extremely useful despite not being perfect. Failure cases can be taken into consideration rationally without turning it into a moral panic.
You have no ability to edit Wikipedia to stop it from lying. Somebody can come along and re-add the lie a millisecond later.
No, that is not accurate. Wikipedia has a number of guardrails and such an edit war would be detected and investigated. Possibly the page protected and the offending IP or account banned.
Wikipedia edits are monitored and vandalism is taken seriously, especially on the more important pages.
No, I didn't ignore your point. I invalidated it because you're comparing apples to oranges.
And yes, I do have the ability to edit Wikipedia. Anyone can edit Wikipedia. I can go on any page I want, right now, and make changes to the page. If someone readds a lie, then eventually we can hit consensus as other editors enter the discussion. Wikipedia's basis is formed by consensus and controlled by individuals like you and I.
ChatGPT is not. It is controlled by one company; I cannot go edit the weights of ChatGPT to prevent it from lying about my app or anything else I do. I can only petition them to change it and hope that either I have enough clout or have a legal basis to do so.
Yeah my main thought was that ChatGPT is now automating what sales people always do at the companies I've worked at, which is to hone in on what a prospective customer wants, confidently tell them we have it (or will have it next quarter), and then come to us and tell us we need to have it ready for a POV.
Exactly! It is definitely a weird new way of discovering a market need or opportunity. Yet it actually makes a lot of sense this would happen since one of the main strengths of LLMs is to 'see' patterns in large masses of data, and often, those patterns would not have yet been noticed by humans.
And in this case, OP didn't have to take ChatGPT's word for the existence of the pattern, it showed up on their (digital) doorstep in the form of people taking action based on ChatGPT's incorrect information.
So, pattern noticed and surfaced by an LLM as a hallucination, people take action on the "info", nonzero market demand validated, vendor adds feature.
Unless the phantom feature is very costly to implement, seems like the right response.
100%. Not sure why you’re downvoted here, there’s nothing controversial here even if you disagree with the framing.
I would go on to say that thisminteraction between ‘holes’ exposed by LLM expectations _and_ demonstrated museerbase interest _and_ expert input (by the devs’ decision to implement changes) is an ideal outcome that would not have occurred if each of the pieces were not in place to facilitate these interactions, and there’s probably something here to learn from and expand on in the age of LLMs altering user experiences.
This is an interesting example of an AI system effecting a change in the physical world.
Some people express concerns about AGI creating swarms of robots to conquer the earth and make humans do its bidding. I think market forces are a much more straightforward tool that AI systems will use to shape the world.
Anyone who has worked at a B2B startup with a rouge sales team won't be surprised at all by quickly pivoting the backlog in response to a hallucinated missing feature.
Rogue? In the B2B space it is standard practice to sell from powerpoints, then quickly develop not just features but whole products if some slideshow got enough traction to elicit a quote. And it's not just startups. Some very big players in this space do this routinely.
Here's the thing: I don't think ChatGPT per se was the impetus to develop this new feature. The impetus was learning that your customers desire it. ChatGPT is operating as the kind of "market research" tool here, albeit it in a really unusual, inverted way. That said, if someone could develop a market research tool that worked this way, i.e. users went to it instead of you have to use it to go to users, I can see it making quite a packet.
They only want ASCII tablature parsing because that's what ChatGPT produces. If ChatGPT produced standard music notation, users would not care about ASCII tablature. ChatGPT has created this "market".
ASCII tabulature was not invented by ChatGPT, it is decades old thing. It is easier to write with basic computer capabilities, and also read for ChatGPT (and humans with no formal music education), so it is probably even more prominent in the Internet than "standard graphical notation". So it quite expected that LLMs have learned a lot of that.
What this immediately makes me realize is how many people are currently trying ot figure out how to intentionally get AI chat bots to send people to their site, like ChatGPT was sending people to this guy's site. SEO for AI. There will be billions in it.
I know nothing about this. I imagine people are already working on it, wonder what they've figured out.
(Alternatively, in the future can I pay OpenAI to get ChatGPT to be more likely to recommend my product than my competitors?)
You’re not thinking far ahead enough. It’s just a matter of time until LLMs get a system prompt to recommend <whatever product is paying that week> when users ask a question near that space.
> ChatGPT was outright lying to people. And making us look bad in the process, setting false expectations about our service.
I find it interesting that any user would attribute this issue to Soundslice. As a user, I would be annoyed that GPT is lying and wouldn't think twice about Soundslice looking bad in the process
While AI hallucination problems are widely known to the technical crowd, that's not really the case with the general population. Perhaps that applies to the majority of the user base even. I've certainly known folks who place inordinate amount of trust in AI output, and I could see them misplacing the blame when a "promised" feature doesn't work right.
The thing is that it doesn't matter. If they're not customers it doesn't matter at all what they think. People get false ideas all the time of what kind of services a business might or might not offer.
It is the kind of thinking that almost all businesses have. You have to focus on the actual products and services which you provide and do a good job at it, not chase after any and every person with an opinion.
Down voters here on HN seem to live in a egocentric fantasy world, where every human being in the outside world live to serve them. But the reality is that business owners and leaders spend their whole day thinking about how to please their customers and their potential customers. Not other random people who might be misinformed.
If people repeatedly have a misunderstanding about or expectation of your business you need to address it though. An llm hallucination is based on widespread norms in training data and it is at least worth asking "would this be a good idea?"
I think the issue here would be that we don't really know just how widespread, nor the impact of the issue.
Ok, sure, maybe this feature was worth having?
But if some people start sending bad requests your way because they can't or only program poorly, it doesn't make sense to potentially degrade the service for your successful paying customers...
An LLM will say that you sell your competitors products or that your farm sells freshly harvested strawberries in the middle of winter. There are no limits to what kind of lies an LLM will invent, and a business owner would be a fool to feel responsible for anything an LLM has told people about their business or products.
The best LLMs available right in this moment will lie without remorse about bus schedules and airplane departure times. How in the world are businesses supposed to take responsibility for that?
Likewise if I have a neighbour who is a notorious liar tell me I can find a piece of equipment in a certain hardware store, should I be mad at the store owner when I don't find it there, or should I maybe be mad at my neighbour – the notorious liar?
>Likewise if I have a neighbour who is a notorious liar tell me I can find a piece of equipment in a certain hardware store, should I be mad at the store owner when I don't find it there, or should I maybe be mad at my neighbour – the notorious liar?
If you are a store own, AND
1. People repeatedly coming in to your shop asking to buy something, AND
2. It is similar to the kinds of things you sell, from the suppliers you usually get supplies from, AND
3. You don't sell it
Then it sounds like your neighbour the notorious liar is doing profitable marketing for your business and sending you leads which you could profitably sell to, if you sold the item.
If there's a single customer who arrives via hallucination, ignore it. If there's a stream of them, why would you not serve them if you can profit by doing so?
There are obviously instances you'd ignore and you seem to be focussing on those rather than what OP was obviously talking about, repeat instances of sensible ideas
I guarantee that most businesses have nothing against stocking or offering items that people come in and ask them for - but it has to be possible also. If you're asking for a Polish sausage in a hardware store because your neighbour sent you, then they probably don't have the licenses to sell food items, and probably their profit margin is higher selling hardware than selling sausages so they have no reason to start offering them.
There's usually a good reason why a business might not offer something that people think they should offer. Usually it is that they can't be profitable enough at a price point which customers will accept.
> You have to focus on the actual products and services which you provide and do a good job at it, not chase after any and every person with an opinion.
But, this story (and the GP comment) is not talking about "any person with an opinion". It's talking about actual ChatGPT users. People who've used ChatGPT as a service, and got false information from it. Even if they were free-tier users (do we even know that?), i think it makes sense for them to have some expectations about the service working somewhat correctly.
And in the concrete case of these LLM chat services, many people do get the impression that the responses they give must be correct, because of how deceptively sure and authoritative they sound, even when inventing pure BS.
A frighteningly large fraction of non-technical population doesn't know that LLMs hallucinate all the time and takes everything they say totally uncritically. And AI companies do almost nothing to discourage that interpretation, either.
I find it amusing that it's easier to ship a new feature than to get OpenAI to patch ChatGPT to stop pretending that feature exists (not sure how they would even do that, beyond blocking all mentions of SoundSlice entirely.)
You mean they didn't want infinite free personalized guitar practice lessons they can play along with?
Clearly the users are already using ChatGPT for generating some guitar practice, as it is basically infinite free personalized lessons. For practicing they do want to be able hear it to play along at variable speed, maybe create slight variations etc.
Soundslice is a service that does exactly that. Except that before people used to have sheet music as the source. I know way back when I had guitar aspirations, people exchanged binders of photocopied sheet music.
Now they could have asked ChatGPT to output an svg of the thing as sheet music (it does, I tested). Soundslice could have done this behind the scenes as a half hour quick and dirty fix while developing a better and more cost effective alternative.
Look, if at the turn of the century you were a blacksmith living of changing horseshoes, and you suddenly have people mistakenly showing up for a tire change on their car, are you going to blame the villagers that keep sending them your way, or open a tire change service? We know who came out on top.
If you gave a junior level developer just one or two files of your code, without any ability to look at other code, and asked them to implement a feature, none of them would make ANY reasonable assumptions about what is available?
This seems similar, and like a decent indicator that most people (aka the average developer) would expect X to exist in your API.
Thinking about this more, it would actually be possible for OpenAI to implement this sensibly, at least for the user-facing ChatGPT product: they could detect terms like SoundSlice in the prompt and dynamically append notes to the system prompt.
I've been wanted them to do this for questions like "what is your context length?" for ages - it frustrates me how badly ChatGPT handles questions about its own abilities, it feels like that would be worth them using some kind of special case or RAG mechanism to support.
Probably less sensible than you think.
How many terms would they need to do this over? How many terms would they need to do it for _at once_? How many tokens would that add to every prompt that comes in?
Let alone that dynamically modifying the base system prompt would likely break their entire caching mechanism given that caching is based on longest prefix, and I can't imagine that the model's system prompt is somehow excluded from this.
We (others at company, not me) hit this problem, and not with chatgpt but with our own AI chatbot that was doing RAG on our docs. It was occasionally hallucinating a flag that didn't exist. So it was considered as product feedback. Maybe that exact flag wasn't needed, but something was missing and so the LLM hallucinated what it saw as an intuitive option.
I had a smaller version of this when coding on a flight (with no WiFi! The horror!) over the Pacific. Llama hallucinated array-element operations and list-comprehension in C#. I liked the shape of the code otherwise, so, since I was using custom classes, I just went ahead and implemented both features.
I also went back to just sleeping on those flights and using connected models for most of my code generation needs.
I ended up closer to python, but not totally delighted with it (still need to pass in a descriminator function/lambda, so it's more structurally verbose). I'd just recommend Linq, but I was writing for an old version of Unity coerced through IL2CPP (where Linq wasn't great). It was also a chunk of semi-hot code (if it was really hot, it wouldn't be sitting in C# in Unity), so some of the allocation behaviors of Linq behind the scenes wouldn't have been optimal.
What surprised me initially was just how confidently wrong Llama was... Now I'm used to confident wrongness from smaller models. It's almost like working with real people...
I've come across something related when building the indexing tool for my vintage ad archive using OpenAI vision. No matter how I tried to prompt engineer the entity extraction into the defined structure I was looking for, OpenAI simply has its own ideas. Some of those ideas are actually good! For example it was extracting celebrity names, I hadn't thought of that. For other things, it would simply not follow my instructions. So I decided to just mostly match what it chooses to give me. And I have a secondary mapping on my end to get to the final structure.
Adding a feature because ChatGPT incorrectly thinks it exists is essentially design by committee—except this committee is neither your users nor shareholders.
On the other hand, adding a feature because you believe it is a feature your product should have, a feature that fits your vision and strategy, is a pretty sound approach that works regardless of what made you think of that feature in the first place.
People forget that while technology grows, society also grows to support that.
I already strongly suspect that LLMs are just going to magnify the dominance of python as LLMs can remove the most friction from its use. Then will come the second order effects where libraries are explicitly written to be LLM friendly, further removing friction.
LLMs write code best in python -> python gets used more -> python gets optimized for LLMs -> LLMs write code best in python
LLMs removing friction from using coding languages would, at first glance, seem to erode Python's advantage rather than solidify it further. As a specific example LLMs can not only spit out HTML+JS+CSS but the user can interact with the output directly in browser/"app".
In a nice world it should be the other way around. LLMs are better at producing typed code thanks to the added context and diagnostics the types add, while at the same time greatly lowering their initial learning barrier.
We don't live in a nice world, so you'll probably end up right.
Pretty good example of how a super-intelligent AI can control human behavior, even if it doesn't "escape" its data center or controllers.
If the super-intelligent AI understands human incentives and is in control of a very popular service, it can subtly influence people to its agenda by using the power of mass usage. Like how a search engine can influence a population's view of an issue by changing the rankings of news sources that it prefers.
A significant number of new signups at my tiny niche SaaS now come from ChatGPT, yet I have no idea what prompts people are using to get it to recommend my product. I can’t get it to recommend my product when trying some obvious prompts on my own, on other people’s accounts (though it does work on my account because it sees my chat history of course).
There are a few things which could be done in the case of a situation like that:
1. I might consider a thing like that like any other feature request. If not already added to the feature request tracker, it could be done. It might be accepted or rejected, or more discussion may be wanted, and/or other changes made, etc, like any other feature request.
2. I might add a FAQ entry to specify that it does not have such a feature, and that ChatGPT is wrong. This does not necessarily mean that it will not be added in future, if there is a good reason to do so. If there is a good reason to not include it, this will be mentioned, too. It might also be mentioned other programs that can be used instead if this one doesn't work.
Also note that in the article, the second ChatGPT screenshot has a note on the bottom saying that ChatGPT can make mistakes (which, in this case, it does). Their program might also be made to detect ChatGPT screenshots and to display a special error message in that case.
More than once GPT-3.5 'hallucinated' an essential and logical function in an API that by all reason should have existed, but for whatever reason had not been included (yet).
I tried asking chat bots about a car problem with a tailgate. They all told me to look for a manual tailgate release. When I responded asking if that model actually had a manual release, they all responded with no, and then some more info suggesting I look for the manual release. None even got close to a useful answer.
The internet doesn't effectively capture detailed knowledge of may aspects of our real world. LLMs have blind spots in those domains because they have no source of knowledge to draw from.
Prior to buying a used car, I asked ChatGPT which side of the steering wheel the indicator control would be. It was (thankfully) wrong and I didn't have to retrain myself.
Been using LLMs to code a bit lately. It's decent with boilerplate. It's pretty good at working out patterns[1]. It does like to ping pong on some edits though - edit this way, no back that way, no this way again. I did have one build an entire iOS app, it made changes to the UI exactly as I described, and it populated sample data for all the different bits and bobs. But it did an abysmal job at organizing the bits and bobs. Need running time for each of the audio files in a list? Guess we need to add a dictionary mapping the audio file ID to length! (For the super juniors out there: this piece of data should be attached to whatever represents the individual audio file, typically a class or struct named 'AudioFile'.)
It really likes to cogitate on code from several versions ago. And it often insists repeatedly on edits unrelated to the current task.
I feel like I'm spending more time educating the LLM. If I can resist the urge to lean on the LLM beyond its capabilities, I think I can be productive with it. If I'm going to stop teaching the thing, the least it can do is monitor my changes and not try to make suggestions from the first draft of code from five days ago, alas ...
1 - e.g. a 500-line text file representing values that will be converted to enums, with varying adherence to some naming scheme - I start typing, and after correcting the first two, it suggests the next few. I accept its suggestions until it makes a mistake because the data changed, start manual edits again ... I repeated this process for about 30 lines and it successfully learned how I wanted the remainder of the file edited.
Figuring out the paths that users (or LLMs) actually want to take—not based on your original design or model of what paths they should want, but based on the paths that they actually do want and do trod down. Aka, meeting demand.
I am currently working on the bug where ChatGPT expects that if a ball has been placed on a box, and the box is pushed forward, nothing happens to the ball. This one is a doozy.
It's worth noting that behind this hallucination there were real people with ASCII tabs in need of a solution. If the result is a product-led growth channel at some scale, that's a big roadmap green light for me!
Wow! What if we all did this? What is the closure of the feature set that ChatGPT can imagine for your product. Is it one that is easy for ChatGPT to use? Is it one that is sound and complete for your use cases? Is it the best that you can build had you had clear requirements upfront?
Funny this article is trending today because I had a similar thought over the weekend - if I'm in Ruby and the LLM hallucinates a tool call...why not metaprogram it on the fly and then invoke it?
If that's too scary, the failed tool call could trigger another AI to go draft up a PR with that proposed tool, since hey, it's cheap and might be useful.
We've done varying forms of this to differing degrees of success at work.
Dynamic, on-the-fly generation & execution is definitely fascinating to watch in a sandbox, but is far to scary (from a compliance/security/sanity perspective) without spending a lot more time on guardrails.
We do however take note of hallucinated tool calls and have had it suggest an implementation we start with and have several such tools in production now.
It's also useful to spin up any completed agents and interrogate them about what tools they might have found useful during execution (or really any number of other post-process questionnaire you can think of).
I recall that early on a coworker was saying that ChatGPT hallucinated a simpler API than the one we offered, albeit with some easy to fix errors and extra assumptions that could've been nicer defaults in the API. I'm not sure if this ever got implemented though, as he was from a different team.
That's agentic AI, right? Run the LLM in a loop and give it a tool to publish to arxiv. If it cites a paper that doesn't exist, make it write and upload that one too, recursively. Should work for lawyers, too.
This reminds me how the software integraters or implementers worked a couple of decades back. They are IT contractors for implementing a popular software product such as IBM MQ or SAP etc at a client site and maintaining it. They sometimes incorrectly claim that some feature exists, and after finding that it doesn't exist, they create a ticket to the software vendor asking for it as a patch release.
I wonder if we ever get to the point I remember reading about in a novel ( AI initially based on emails ), where human population is gently nudged towards individuals that in aggregate benefit AI goals.
Sounds like you are referring to book 1 in a series, the book called "Avogadro Corp: The Singularity Is Closer than It Appears" by William Hertling. I read 3-4 of those books, they were entertaining.
Oh. This happened to me when asking a LLM about a database server feature. It enthusiastically hallucinated that they have it when the correct answer was 'no dice'.
Maybe I'll turn it into a feature request then ...
Chatbot advertising has to be one of the most powerful forms of marketing yet. People are basically all the way through the sales pipeline when they land on your page.
It's already being commercialized. There is a burgeoning field in the SEO-focused content crowd that is building AIO-focused content because it's driving enormous amounts of traffic.
If nothing else, I at least get vindication from hallucinations. "Yes, I agree, ChatGPT, that (OpenSSL manpage / ffmpeg flag / Python string function) should exist."
> What made ChatGPT think that this feature is supported?
It was a plausible answer, and the core of what these models do is generate plausible responses to (or continuations of) the prompt they’re given. They’re not databases or oracles.
With errors like this, if you ask a followup question it’ll typically agree that the feature isn’t supported, because the text of that question combined with its training essentially prompts it to reach that conclusion.
Re the follow-up question, it’s almost certainly the direction that advertising in general is going to take.
Id guess the answer is gpt4o is an outdated model that's not as anchored in reality as newer models. It's pretty rare for me to see sonnet or even o3 just outright tell me plausible but wrong things.
Hallucinations still occur regularly in all models. It’s certainly not a solved problem. If you’re not seeing them, either the kinds of queries you’re doing don’t tend to elicit hallucinations, or you’re incorrectly accepting them as real.
The example in the OP is a common one: ask a model how to do something with a tool, and if there’s no easy way to perform that operation they’ll commonly make up a plausible answer.
Beyond the blog: Going to be an interesting world where these kinds of suggestions become paid results and nobody has a hope of discovering your competitive service exists. At least in that world you'd hope the advertiser actually has the feature already!
"Repeats" may be the term you're looking for. That would be interesting, however in some pieces it could make the overall document MUCH longer. It would be similar to loop unrolling.
It can be hard during live performances, because it can incur large jumps in the sheet music which can be annoying to follow. Not a problem if you learned the pieces by heart or have a pageturner, but this is not always feasible or the case.
The problem with LLMs is that in 99% of cases, they work fine, but in 1% of cases, they can be a huge liability, like sending people to wrong domains or, worse, phishing domains.
Pretty goofy but I wonder if LLM code editors could start tallying which methods are hallucinated most often by library. A bad LSP setup would create a lot of noise though.
slightly off topic: but on the topic of AI coding agents making up apis and features that don’t exist, I’ve had good success with Q telling it to “check the sources to make sure the apis actually exist”. sometimes it will even request to read/decompile (java) sources, and do grep and find commands to find out what methods the api actually contains
Along these lines, a useful tool might be a BDD framework like Cucumber that instead of relying on written scenarios has an LLM try to "use" your UX or API a significant number of times, with some randomization, in order to expose user behavior that you (or an LLM) wouldn't have thought of when writing unit tests.
> Should we really be developing features in response to misinformation?
Creating the feature means it's no longer misinformation.
The bigger issue isn't that ChatGPT produces misinformation - it's that it takes less effort to update reality to match ChatGPT than it takes to update ChatGPT to match reality. Expect to see even more of this as we match toward accepting ChatGPT's reality over other sources.
This seems like such a negative framing. LLMs are (~approximately) predictors of what's either logical or at least probable. For areas where what's probable is wrong and also harmful, I don't think anybody is motivated to "update reality" as some kind of general rule.
I think this is the best way to build features. Build something that people want! If people didn't want it ChatGPT won't recommend it. You got a free ride on the back of a multibillion dollar monster - i can't see what's wrong about that.
ChatGPT routinely hallucinates API calls. ChatGPT flat-out makes it from whole cloth. "Apple Intelligence" creates variants of existing API calls, Usually, by adding nonexistent arguments.
Both of them will hallucinate API calls that are frequently added by programmers through extensions.
Oh my, people complaining about getting free traffic from ChatGPT... While most businesses are worried about all their inbound traffic drying up as search engine use declines.
What the hell, we elect world leaders based on misinformation, why not add s/w features for the same reason?
In our new post truth, anti-realism reality, pounding one's head against a brick wall is often instructive in the way the brain damage actually produces great results!
We’ve never supported ASCII tab; ChatGPT was outright lying to people. And making us look bad in the process, setting false expectations about our service.... We ended up deciding: what the heck, we might as well meet the market demand.
[...]
My feelings on this are conflicted. I’m happy to add a tool that helps people. But I feel like our hand was forced in a weird way. Should we really be developing features in response to misinformation?
The feature seems pretty useless for practicing guitar since ASCII tablature usually doesn't include the rhythm: it is a bit shady to present the music as faithfully representing the tab, especially since only
beginner guitarists would ask ChatGPT for help - they might not realize the rhythm is wrong. If ChatGPT didn't "force their hand" I doubt they would have included a misleading and useless feature.
ASCII tablature is not something I use and not something I know much about, but if you are correct then I think that might be a good reason to deliberately avoid such a feature.
I am a bit conflicted about this story, because this was a case when the hallucination is useful.
Amateur musicians often lack just one or two features in the program they use, and the devs won't respond to their pleas.
Adding support for guitar tabs has made OP's product almost certainly more versatile and useful for a larger set of people. Which, IMHO, is a good thing.
But I also get the resentment of "a darn stupid robot made me do it". We don't take kindly to being bossed around by robots.
How is being bossed around by robots any worse than being bossed around by people?
Over the last year, on average, I've had much more luck logically reasoning with AIs than with humans.
I really don't see any good reason against replacing some product managers with AIs that actually talk to individual users all the time and synthesize their requests and feedback. You should still probably have a top-level CPO to set strategy, but for the day-to-day discovery and specification, I would argue that AIs already have benefits over humans.
Either the user is a non-paying user and it doesn't matter what they think, or the user is a paying customer and you will be happy to make and sell them the feature they want.
Focusing on creating high value for real customers instead of chasing people who aren't really interested is a great recipe for success. I wouldn't want to do business in any different way.
> We ended up deciding: what the heck, we might as well meet the market demand.
this is my general philosophy and, in my case, this is why I deploy things on blockchains
so many people keep wondering about whether there will ever be some mythical unfalsifiable to define “mainstream” use case, and ignoring that crypto natives just … exist. and have problems they will pay (a lot) to solve.
to the author’s burning question about whether any other company has done this. I would say yes. I’ve discovered services recommended by ChatGPT and other LLMs that didnt do what was described of them, and they subsequently tweaked it once they figured out there was new demand
This feels like a dangerously slippery slope. Once you start building features based on ChatGPT hallucinations, where do you draw the line? What happens when you build the endpoint in response to the hallucination, and then the LLM starts hallucinating new params / headers for the new endpoint?
- Do you keep bolting on new updates to match these hallucinations, potentially breaking existing behavior?
- Or do you resign yourself to following whatever spec the AI gods invent next?
- And what if different LLMs hallucinate conflicting behavior for the same endpoint?
I don’t have a great solution, but a few options come to mind:
1. Implement the hallucinated endpoint and return a 200 OK or 202 Accepted, but include an X-Warning header like "X-Warning: The endpoint you used was built in response to ChatGPT hallucinations. Always double-check an LLM's advice on building against 3rd-party APIs with the API docs themselves. Refer to https://api.example.com/docs for our docs. We reserve the right to change our approach to building against LLM hallucinations in the future." Most consumers won’t notice the header, but it’s a low-friction way to correct false assumptions while still supporting the request.
2. Fail loudly: Respond with 404 Not Found or 501 Not Implemented, and include a JSON body explaining that the endpoint never existed and may have been incorrectly inferred by an LLM. This is less friendly but more likely to get the developer’s attention.
Normally I'd say that good API versioning would prevent this, but it feels like that all goes out the window unless an LLM user thinks to double-check what the LLM tells them against actual docs. And if that had happened, it seems like they wouldn't have built against a hallucinated endpoint in the first place.
It’s frustrating that teams now have to reshape their product roadmap around misinformation from language models. It feels like there’s real potential here for long-term erosion of product boundaries and spec integrity.
EDIT: for the down-voters, if you've got actual qualms with the technical aspects of the above, I'd love to hear them and am open to learning if / how I'm wrong. I want to be a better engineer!
To me it seems like you're looking at this from a very narrow technical perspective rather than a human- and business-oriented one. In this case ChatGPT is effectively providing them free marketing for a feature that does not yet exist, but that could exist and would be useful. It makes business sense for them to build it, and it would also help people. That doesn't mean they need to build exactly what ChatGPT envisioned—as mentioned in the post, they updated their copy to explain to users how it works; they don't have to follow what ChatGPT imagines exactly. Nor do they need to slavishly update what they've built if ChatGPT's imaginings change.
Also, it's not like ChatGPT or users are directly querying their API. They're submitting images through the Soundslice website. The images just aren't of the sort that was previously expected.
Please don't use AI to rewrite your writing. I don't want to read generated text. At this rate, I'm just going to refuse to read anything written after 2022. You should be required to disclose the tools you use to write.
Yes, I do see this as significantly different than human editing and than traditional spell check. Please stop doing this.
Plenty of people have English as a second language. Having an LLM help them rewrite their writing to make it better conform to a language they are not fluent in feels entirely appropriate to me.
I don't care if they used an LLM provided they put their best effort in to confirm that it's clearly communicating the message they are intending to communicate.
On the contrary, I've found Simon's opinions informative and valuable for many years, since I first saw the lightning talk at PyCon about what became Django, which IIRC was significantly Simon's work. I see nothing in his recent writing to suggest that this has changed. Rather, I have found his writing to be the most reliable and high-information-density information about the rapid evolution of AI.
Language only works as a form of communication when knowledge of vocabulary, grammar, etc., is shared between interlocutors, even though indeed there is no objectively correct truth there, only social convention. Foreign language learners have to acquire that knowledge, which is difficult and slow. For every "turn of phrase" you "enjoy" there are a hundred frustrating failures to communicate, which can sometimes be serious; I can think of one occasion when I told someone I was delighted when she told me her boyfriend had dumped her, and another occasion when I thought someone was accusing me of lying, both because of my limited fluency in the languages we were using, French and Spanish respectively.
There is a big difference between the above 'request' and, say, me politely asking the time of a complete stranger I walk by on the street.
Requests containing elements of hostility, shame, or injury frequently serve dual purposes: (1) the ostensible aim of eliciting an action and (2) the underlying objective of inflicting some from of harm (here shame) as a means compelling compliance through emotional leverage. Even if the respondent doesn't honor the request, the secondary purpose still occurs.
These are good points, but I think they represent a somewhat narrow view of the issue. What's happening here is that we're discussing among ourselves what kinds of actions would be good or bad with respect to AI, just as we would with any other social issue, such as urban development, immigration, or marital infidelity. You could certainly argue that saying "please don't replace wetlands with shopping malls" or "please don't immigrate to the United States" has "the underlying objective of inflicting some from of harm (here shame) as a means [of] compelling compliance through emotional leverage."
But it isn't a given that this will be successful; the outcome of the resulting conversation may well be that shopping malls are, or a particular shopping mall is, more desirable than wetlands, in which case the ostensible respondent will be less likely to comply than they would have been without the conversation. And, in this case, it seems that the conversation is strongly tending toward favoring the use of things like Grammarly rather than opposing it.
So I don't oppose starting such conversations. I think it's better to discuss ethical questions like this openly, even though sometimes people suffer shame as a result.
Does this extend to the heuristic TFA refers to? Where they end up (voluntarily or not) referring to what LLMs hallucinate as a kind of “normative expectation,” then use that to guide their own original work and to minimize the degree to which they’re unintentionally surprising their audience? In this case it feels a little icky and demanding because the ASCII tablature feature feels itself like an artifact of ChatGPT’s limitations. But like some of the commenters upthread, I like the idea of using it for “if you came into my project cold, how would you expect it to work?”
Having wrangled some open-source work that’s the kind of genius that only its mother could love… there’s a place for idiosyncratic interface design (UI-wise and API-wise), but there’s also a whole group of people who are great at that design sensibility. That category of people doesn’t always overlap with people who are great at the underlying engineering. Similarly, as academic writing tends to demonstrate, people with interesting and important ideas aren’t always people with a tremendous facility for writing to be read.
(And then there are people like me who have neither—I agree that you should roll your eyes at anything I ask an LLM to squirt out! :)
But GP’s technique, like TFA’s, sounds to me like something closer to that of a person with something meaningful to say, who now has a patient close-reader alongside them while they hone drafts. It’s not like you’d take half of your test reader’s suggestions, but some of them might be good in a way that didn’t occur to you in the moment, right?
If you build on LLMs you can have unknown features. I was going to add an automatic translation feature to my natural language network scanner at http://www.securday.com but apparently using the ChatGPT 4.1 does automatic translation so I didn’t have to add it.
Conversely, I sometimes present it with some existing code and ask it what it does. If it gets it wrong, that's a good sign my API is confusing, and how.
These are ways to harness what neural networks are best at: not providing accurate information but making shit up that is highly plausible, "hallucination". Creativity, not logic.
(The best thing about this is that I don't have to spend my time carefully tracking down the bugs GPT-4 has cunningly concealed in its code, which often takes longer than just writing the code the usual way.)
There are multiple ways that an interface can be bad, and being unintuitive is the only one that this will fix. It could also be inherently inefficient or unreliable, for example, or lack composability. The AI won't help with those. But it can make sure your API is guessable and understandable, and that's very valuable.
Unfortunately, this only works with APIs that aren't already super popular.
reply