I work in (okay, adjacent to) finance. Any communications that are sent / made available to people outside your own organisation are subject to being interpreted as legally binding to various degrees. Provenance of any piece of text/diagram is vitally important.
Let's pair this with a real life example: Google's Gemini sales team haven't understood the above. Their splashy sales pitch for using Gemini as part of someone's workflow is that it can autogenerate document sections and slide decks. The idea of annotating sections based on whether they were written by a human or an unaccountable tool appeared entirely foreign to them.
(The irony is that Google would be particularly well placed to have such annotations. Considering the underlying data structures are CRDTs, and they already show who made any given edit, including an annotation whether the piece of content came from a human or bot should be relatively easy.)
I don't understand this argument. There is accountability: the user or management is always possible to blame.
Say one of my tasks is writing a document, I use a LLM and it tells people to eat rat poison.
But I'm accountable to my boss. My boss doesn't care a LLM did it, my boss cares I submitted something that horrible as completed work.
And if my boss lets that through then my boss is accountable to their boss.
And if my company posts that on the website, then my company is accountable to the world.
Annotations would be useful, sure. But I don't think for one minute they'd release you from any liability. Maybe they don't make it into the final PDF. Or maybe not everyone understands what they're supposed to take away from them. You post it, you'll be held responsible.
Hm, we may be using the word in slightly different tones then. For me accountability is more than just appointing blame, it's also about how you got to the result you brought out.
On the other hand, I absolutely agree with this:
> And if my company posts that on the website, then my company is accountable to the world.
I take pride in having my name associated with material we post publicly. It doesn't make my employer any less involved in it, but it does mean we both put out necks out. The company figuratively, and me personally.
<< my boss cares I submitted something that horrible as completed work.
Bosses come in many shapes and sizes. That said, some of the bosses I had usually wanted it all ( as in: LLM speed, human insights, easy to read format, but also good and complete artifact for auditors ). And they tended to demand it all ( think Musk ) as a way of managing, because they think it helps people work at their highest potential.
Ideally, yes, sadly examples abound with excuses like "the machine did it" or "the machine doesn't seem to allow me to do what you are asking for due to either my own incompetence, those of engineers that fabricated it it, or my organization's policy, so I'm going to pretend it's impossible (even though it would be possible to do it by hand)".
I work in (okay, adjacent to) finance. Any communications that are sent / made available to people outside your own organisation are subject to being interpreted as legally binding to various degrees. Provenance of any piece of text/diagram is vitally important.
Let's pair this with a real life example: Google's Gemini sales team haven't understood the above. Their splashy sales pitch for using Gemini as part of someone's workflow is that it can autogenerate document sections and slide decks. The idea of annotating sections based on whether they were written by a human or an unaccountable tool appeared entirely foreign to them.
(The irony is that Google would be particularly well placed to have such annotations. Considering the underlying data structures are CRDTs, and they already show who made any given edit, including an annotation whether the piece of content came from a human or bot should be relatively easy.)