Just listened to Vogue on my headphones, and I'm unable to notice anything unique about the soundscape of this song. What do you folks hear and when do you hear it? At what point in the song?
That makes sense, given the description of how it works, but the article distinctly mentions getting this effect from gaming headphones so...
Anyway, I went and tried again with just my laptop (based on the person above effusing about it). Again, for me, I'm not hearing anything special, and nothing "3D" about the sounds, other than some left-right shifts. And I guess the music is in front of me, since that's where my laptop is ;)
I don't know about Madonna, but the experience I had listening to this song for the first time with headphones on is what I would guess to be a similar experience:
This is the first time I've ever really noticed a difference between the upload and the CD. The CD Audio glides around your head like a halo, on the YT vid it jitters.
I can hear it distinctly from most of the tracks in the song, especially the voice tracks and some of the sporadic synths. Listening wired on decent headphones if it makes any difference. There's a clear spacial separation that isn't just this thing panned slightly left and that thing panned slightly right.
I agree. The article mentions bits going around their head, or in the front-right. I hear things on the left, and things on the right. And stuff that appears on the opposite side of where the article mentions them. I’m listening on Spotify, maybe all that stuff is stripped out?
Yeah, a 7B foundation model is of course going to be worse when expected to perform on every task.
But finetuning on just a few tasks?
Depending on the task, it's totally reasonable to expect that a 7B model might eke out a win against stock GPT4. Especially if there's domain knowledge in the finetune, and the given task is light on demand for logical skills.
I agree, I think they need an example or two on that blog post to back up the claim. I'm ready to believe it, but I need something more than "diverse customer tasks" to understand what we're talking about.
You can fine-tune a small model yourself and see. GPT-4 is an amazing general model, but won’t perform the best at every task you throw at it, out of the box. I have a fine-tuned Mistral 7B model that outperforms GPT 4 on a specific type of structured data extraction. Maybe if I fine-tuned GPT-4 it could beat it, but that costs a lot of money for what I can now do locally for the cost of electricity.
Not for translations. Did a lot of experimenting different local models. None come even a bit close to the capabilities of chatgpt. Most local models just outputting plain wrong intormation. I am still hoping one day it will be possible. For our business a huge opportunity.
Hmm. Says "shift-click" to open poster on NeurIPS site -- you'll have to be logged in to the NeurIPS virtual conference, and the authors will have had to remember to upload their poster PDF.
If you want the paper pdf, you can "alt-click" to open paper on OpenReview.
The author claims that Van Eyck notices details we'd otherwise miss out and uses the Arnolfini portrait as supporting evidence. However it's possible that artifacts such as the beads appear not because he saw the world differently, but due to his artistic choice of tools. He may have used an optical projection device, a "perspective machine" -- Which means that his process may not have allowed for skipping the inclusion of the beads.
hoestly that's an real disservice to how hard it would be even to paint something from a projected image, this is the same kind of reductive judgement digital artists get today about making their art, I'd challenge anyone to try and take a camera obscura make a painting that looks as good as that one without the vast experience and training behind even simple things like handling and mixing pigments and filling in details that have moved.
even just knowing what you're looking at is a process of un-training your brain, a good art education actually starts from learning to see "correctly" and unsee all the shortcuts that your brain takes.
Absolutely agree that this is an incredibly hard skill in any medium, digital or otherwise.
It's also possible that Van Eyck may have perceived the world the same way we do, and the inclusion of small details such as the beads came only due to his artistic choice of tools. Conversely if you were trained to the same tool for sketching you would be forced to include the beads by way of process -- even if you had not noticed them before.
Of course, as far as artistry goes, it really doesn’t matter.
When I was first considering getting a BFA (in my late 30s) I thought my inability to photographically reproduce scenes using pencils and other traditional media precluded my going to art school... but I still gathered my courage, front-loaded my portfolio with what I thought was my strongest work— very polished photography— and scheduled a “do I even stand a chance here” portfolio consultation.
I swallowed hard when the interviewer (gruffly) introduced himself as a professor in the fine art photography program. He cooly disregarded shots I thought were most impressive— a perfectly composed and finished shot of a huge lightning bolt over a brick victorian cathedral shot from a fifth floor outdoor vantage point a few blocks away with a full frame DSLR; some long exposure shots of really grand views in a city; a few others. ‘Ah, yes. Very pretty. Doesn’t say much about you though, non?’ But as soon as he got through those and got to the ones I threw in there to pad things out, he started to pay more attention.
One-for-one, he was most interested in the ones I liked the most. I did not even take most on a ‘real’ camera, I took them on my phone... Abstract shapes made by overlapping subway bars. A smashed pile of dropped beer glasses at the bar I worked at. A guy in a suit sitting down waiting for a his hamburger order at 3am. He actually stopped, put his coffee and pen down, and spent a solid minute smiling and just taking in a picture of my wife, from behind, sitting a fifty or so feet away on a beach staring out into the water in the low-left corner of the frame with a low-flying plane the same size as her in top right of the frame. The shot really does viscerally convey groundedness, and airiness, the simultaneous distance and closeness of the juxtaposition on the infinite expanse of ocean and sky... it’s a great, kinda grainy, totally unpolished iPhone photograph.
Without a single pencil portrait or oil painting, he waived my admission and let me in on the spot. My school is no RISD, but a professor who went to RISD prided himself on perfect pencil realism as a high school student, but soon realized how little it mattered in practice. We obviously must know how to produce things visually, but perfectly reproducing a glass sphere sitting in front of a waterfall is a parlor trick. The time spent honing those skills to that level would be much better spend figuring out new and interesting ways to present what you see, or figuring out new angles or new truths to communicate about YOUR world, or even going out and experiencing new things to find captivating about it— many of them you'll probably find on your block.
Many folks— largely folks that haven’t ever or recently had anything to do with formal art education— think we lost something important by deemphasizing those rigid classical technical capabilities. Many of those folks would also be pretty sad if rock, rap, folk, cool jazz, folk, electronica, or any other modern music was replaced by Serious Classical Virtuosity.
In the end, you need exactly as much skill as you need to express what you want to express. The tools or techniques compensating for ‘missing’ skills often become the most compelling elements— like Jan van Eyck, here. He could have gotten a future iPad beamed to him with Adobe Illustrator installed on it and it wouldn’t make those prayer beads (or any of the other captivating shinies in his work) any less stunning. You can instantly tell his work by looking at it and it’s certainly not because everything looks just like a photograph— it has real emotional impact. It really viscerally communicates that weird Van Eck something and it definitely would do a worse job of it had he not used whatever projection doodad he used to get it.
The amount of privacy is configurable, and researchers who would like to tune it (in either direction) can make a plea so the public can decide the amount of privacy loss they're willing give up in order to aid research.
Each culture is different and has different tolerances for privacy and amount of intrusiveness.
This is a standard feature of many browsers, including Safari.
E.g: On OSX you can download a .dmg file or .zip file, and when opening the OS will warn: "XYZ is an application downloaded from the internet. Are you sure you want to open it?". The information about the origin of the file comes from extended attributes.
Windows has the functionality you describe too, but it works only by storing a flag specifying what kind of origin the file has, not specifically what the origin was. Your article seems to indicate that mac OS uses basically the same system as Windows.
EDIT: According to some other comments in this thread, I'm wrong. Mac OS does store the whole origin.
EDIT 2: Looks like I'm wrong about Windows too, which also stores the whole origin. This actually disagrees with what is written in the bug report, so perhaps it needs to be updated.
Apparently they added it back after the initial release.
Support for alternate data streams was initially not implemented in ReFS. In Windows 8.1 64-bit and Server 2012 R2 the file system reacquired support for alternate data streams, with lengths of up to 128K
Additionally, the same information is stored in an SQLite database in your home directory. In fact, the ID at the very end of the output is the primary key to the table:
sqlite3 ~/Library/Preferences/com.apple.LaunchServices.QuarantineEventsV2 "select * from LSQuarantineEvent where LSQuarantineEventIdentifier = 'ADF309D2-762B-4FE2-AEC6-104E019BDBF9'"
The entry in the table also doesn't seem to be deleted when you delete the downloaded file. That is, you can get a list of all files you've ever downloaded:
sqlite3 ~/Library/Preferences/com.apple.LaunchServices.QuarantineEventsV2 "select * from LSQuarantineEvent"
Hmmm, I'm not seeing any entries in that table from Safari (my primary browser). I am seeing all the Hombrew Cask downloads, iChat (messages) and a couple Firefox and Brave based downloads.
> select LSQuarantineDataURLString from LSQuarantineEvent where LSQuarantineAgentName = 'Safari';
https://upload.wikimedia.org/wikipedia/commons/d/d2/Ethiopian_Airlines_ET-AVJ_takeoff_from_TLV_%2846461974574%29.jpg?download
Maybe you've got Safari configured in some way that prevents it, but with the default configuration, Safari definitely gets entries in that db.
Have you modified the configuration of Safari in any way? I never use Safari, except this once to download an image to test with. It seems possible to me that Safari might disable it if you flip some privacy or security switches, which I haven't done.
How many projects would be affected if the Supreme Court sides with Oracle and rules that APIs can't be reimplemented?
Java is open source, but copyright is retained by Oracle. Would this affect say, Amazon's ability to rip off the MongoDB API? What about open source x86 emulators? (Intel IP) Hercules IBM emulator? Any of the Nintendo, SNES etc game emulators?
If Oracle succeeds and Amazon proceeds to sue Oracle for that, that would be some sweet schadenfreude. Granted, I’d still be pissed at the decision, but to have Oracle’s tactics used back at them would feel really nice.
They'd lose for the very same reason that Android is within the confines of the law via the Android Runtime. Dalvik violated the Java license, Android Runtime isn't. Aurora's compatibility layer isn't violating the MySQL license either.
> Dalvik violated the Java license, Android Runtime isn't.
Can you elaborate on this point? What license for what copyrightable work did Dalvik violate, and why does Android Runtime not violate that same copyright license?
The issue at hand was licensing. In the Android Runtime, Google is linking to the OpenJDK libraries rather than using it's "reverse engineered" runtime. OpenJDK is GPL with a classpath exception, which Google is utilizing.
The caveat for Google is any changes to the OpenJDK itself has to be upstreamed. But the Android Runtime itself doesn't have to be GPL due to the classpath exception.
In all honesty, this is how they should have done it in the first place.
It was violating the open source license, thus required a commercial license, which it didn't have, which in turn makes this copyright infringement. At least that's the position of the courts so far.
Linux and OS X implement the same APIs as Unix. Windows implements the same (or very similar) APIs as PM in OS/2.
Most modern PCs reimplement the IBM PC "API", hence "PC compatible".
> In no case does copyright protection for an original work of authorship extend to any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work.
(—Copyright Law of the United States[2])
APIs are the idea that binds components together. Even if you come to understand the interface by reading a programmatic description of it (e.g., a function signature), and reimplement it, and your language forces you to use the same exact wording to match the signature (b/c it's an API), there's nothing different here than the same exception given to the way that a the raw listing of ingredients in a recipe is not copyrightable (and this is called out as an example of such by copyright.gov[1]).
The appellate court erred in its application of the law.
Additionally, this was said on one of the amicus briefs,
> reversing the District Court would dangerously undermine the settled expectations of computer scientists and the entire computer industry that rely upon the open nature of APIs
(—Martin Fowler, Bruce Schneier, Bjarne Stroustrup, et al. were signatories to this amicus brief.)
The decision, were the industry to actually pay attention to it, would wreak havoc on the state of software engineering.
That just means that Apple's implementation is allowed to use the Unix trademark. It really doesn't add much of anything to the copyright discussion, especially since The Open Group doesn't own Unix copyrights, just the trademark.
Yes, if this decision doesn't go Googles way, all hell is going to break lose: WINE, Samba, S3 compatability, Amazon DocumentDB. The list is enormous and frankly, everyone loses somewhere. It's a complete nightmare.
Remember, if Oracle wins, they are not pursuing a ruling that obviates the doctrine of fair use; all of those cover the case of implementing an API for compatibility. Oracle's argument about Android and its use of Java is that Android did not run Java apps; it ran Java code, but it did not implement enough of the Java API surface to let you run Java apps unmodified. You could use existing Java code in new, Android-only apps that couldn't run on Sun/Oracle's JVM.
Therefore, Oracle's argument is that the fair use doctrine does not apply here; that Android is a derivative work of Java, which is not fair use. WINE, for example, is not ruled impermissible by the sort of ruling Oracle seeks.
I agree that Oracle isn't trying to create a ruling that obviates fair use; but assume you believe that the appellate court's ruling is correct, and APIs are copyrightable, how does fair use cover implementing an API for compatibility?
> Oracle's argument about Android and its use of Java is that Android did not run Java apps; it ran Java code, but it did not implement enough of the Java API surface to let you run Java apps unmodified.
So what? Partial compatibility is still meaningful in software engineering, and should allow libraries to run on both. Are you saying that if Google had fully implemented all of the Java APIs, it would have been fair use?
Further, can you ground the above reasoning (that fair use covers API copying, and that Google's partial implementation is material here) in the actual rulings for the case? I don't remember any of this from when I read them, though I do admit that was some time ago.
> WINE, for example, is not ruled impermissible by the sort of ruling Oracle seeks.
Why not? Simply because you cannot write a .exe that would run in WINE, but not in Windows?
Further, what if the example was instead Linux? Linux implements Unix's APIs, but also adds other APIs. You can use existing "Unix" code in new, Linux-only apps that won't run on other *nix OSs.
I would be able to accurately gauge AlphaZero’s true chess ability, strengths and weaknesses. Right now it’s impossible to do that given a curated selection of games. It would be like evaluating someone’s coding ability based on a few samples of their absolute finest work rather than the code they actually write day in and day out.
