dwringer

Kind of funny that the word was capitalized, presumably, due to imitation.

[EDIT: Now changed, so this thread makes less sense, but the original headline was "Artificial intelligence systems found to Excel at ..."]

I would have agreed with your comment 100% after my first visit to such a restaurant, but after my most recent visit I cannot do so. The amount of patience required depends entirely on the quality of service, and this is not necessarily consistent at all such restaurants at all times.

Correct me if I'm wrong but isn't "jailbroken" essentially a term for "used Bayesian techniques to determine what we think somebody probably slipped into the LLM context or trained it on at one point, in theory"?

As I understand you, what you state is exactly what I meant. If you train with a bunch of text containing substrings of those 50-grams, but not the full 50-grams themselves [or, expose it to the same vocabulary used in the same parts of speech as in the full 50], the model will pretty readily produce the full 50-grams despite never having seen them in their entirety. Try it out, it's pretty easy to do on a modern GPU and can be done in less than an hour.

One can fine tune a smaller parameter model like GPT-NeoX on a home GPU pretty readily, and it's absolutely capable of doing what you specified. Teach it with a bunch of example sentences that have parts of speech like verb and noun following a simple grammar, and you will see it generate sentences afterward that combine the parts of speech grammatically in novel ways, using the same grammatical structures but forming productions that did not appear in the training set.

Depending on settings, they are also capable of producing a lot of ungrammatical nonsense, but the odds of what it produces are changed considerably by the training.

Does anyone know what the state of FreeBSD is these days regarding the possible support of CUDA devices? It always pained me that installing it as an OS on my old hardware meant losing access to that, and a large part of the usefulness of the GPU therein.

I've always had great relative pitch but not "perfect pitch"; hearing a note out of the blue without any known context and being able to name it was always a struggle for me, but I did manage to do it first try on this quiz. Different timbres make it easier or harder in my experience. I'm also pretty sure the only reason I can do it now is because with age I got some hearing loss that's not evenly distributed across frequencies so now the difference between different notes is more pronounced.

I know it wasn't the same everywhere, but back in 2001 my high school had an IT staff that consisted of a single teacher with no real technical literacy to begin with. There was no real sort of firewall or security, including the computers that were on the internet. It was kind of a different time.

I'm not sure... hundreds or thousands of code repositories with subtle issues sounds like... the real world of code repositories. And I'd think through analogy and redundancy of some common algorithms, the LLM trained that way might conceivably be able to FIX many of those errors.

There is an extremely weird juxtaposition between the idea of watermarking the outputs so AI-generated samples can be detected if they make up even part of a song (which is arguably considerable as fair use of even copyrighted material if only a single bar is sampled), and the statement:

> Within the experiment, a limited set of creators will be able to use Dream Track for producing a unique soundtrack with the AI-generated voice and musical style of artists including Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Sia, T-Pain, Troye Sivan, and Papoose*.

So... we're watching you and we'll know if you so much as sample one bar to create something of your own. But our preferred creators are having so much fun using AI exclusively to make the entire composition, directly ripping off the personal style of these artists who said it's OK!

I also found it hilarious that they claim the idea of embedding a watermark in the audio by adding the image to the 2d spectrogram is a "novel method ... unlike anything that exists today, especially in the context of audio." Considering Aphex Twin did it in 1999...

* Apparently it was removed from the game, and his Wikipedia doesn't even mention it, but, yes, Papoose, of the "You just mad cuz I'm stylin' on ya!" track from GTA IV.

I too like Windows and use it as my daily OS everywhere, but I am firmly against ever upgrading to Windows 11 and, with each update, Windows 10 gets more unpleasantly like 11.

Recently I tried to open an image, and the Windows Store asked me to pay 99 cents for the codec. Actually, for a video codec pack - it showed me as already having the codec pack for exactly these images installed. But they wouldn't open, only the Windows Store appeared asking me to pay 99 cents. Adding insult to injury, it seems that anyone who did this earlier than a couple of years ago got the pack for free - I was only asked to pay for it because I hadn't secured my rights at the time, apparently. Even though the Windows store verly clearly indicates that I already had the image codec installed.

[even more fun, the Windows store had additional prominent listings for what appeared to be these codecs, for free, but were very transparently third party spyware and not really the codecs at all.]

Windows 11 vs 10 feels in many ways analogous to me to Windows ME vs Windows 2000, at least in terms of user-hostility.

I think "texture" makes sense, if only by analogy to the calligraphic hand Textura, or the etymological derivation of "text" itself.

I agree in part with this philosophy and had success for many years that way, but at some point I started having pain of which simply observing just revealed to me that it was my posture that was partly to blame, at which point I did everything I could to optimize the ergonomics of my workstation. So I conclude that the observational mental approach has merit, but so do ergonomic keyboards and a properly arranged chair and monitor which can pick up where the mental aspect leaves off. At least for me.

Generally making disparaging bigoted comments about a nationality is to be avoided here. It's certainly not productive to this discussion to label anyone as hateful, greedy, and selfish, and it runs afoul of the HN comment guidelines.

There are plenty of ways to contribute to discussion without making remarks that are emotionally charged and inflammatory.

> Before you call my opinion fringe/extreme, go ahead and speak to some low-income folks at your local dive bar.

Now ask them for their opinions on fossil fuels.

Hmm... I do see it, but when I left my comment last night it would seem I had been looking at the grayscale version of the color noise image by mistake. I can still make it out in the blue noise image but it's quite a bit more eyestrain-inducing.

Interestingly, perhaps, I found that one to be easier to make out the fine detail than any other magic eye that I've seen before. However I did find it a bit more eyestrain inducing as well. This is a great blog and makes me want to experiment more.

Almost every post that I read about the subject seems to omit that fact. Ultimately it boils down to a kind of fuzz test and a form of inference over the results. It would be interesting to study the possibilities of quantifying such certainty.

I believe that the application of multiple streams in parallel is a natural evolution of using a single stream. I've used some local models for help in creative writing, and some of the most productive results I got were from running the same prompt and sequence of interactions dozens and dozens of times. Although in that case, I was personally going through each result line by line, I can certainly imagine fully automated tools that leverage the range of responses to a given prompt.

Emacs user here...

Today I learned why sometimes my YAML files seemed inexplicably to end up with tabs in them.

As someone who's worked in the field of used books, I can say from personal experience that's not quite true. It's pretty common to have to type in an ISBN for a variety of reasons. Many times the barcodes have been covered up or defaced, and many publications don't have barcodes in the first place.

I've always felt ligatures that condense two or more glyphs into something that takes up the space of only one in a monospace font are going beyond what a font should handle and into the realm of what an editor should do. I have several such visual substitutions set up in my .emacs but I don't use fonts that do them on their own.

I don't know what I was expecting but that isn't Hurt, it's Surfin' USA with Hurt's lyrics that sound extremely jittery and grainy.

I'm curious though if some AI soon could in fact synthesize the Beach Boys' style with the actual chords and melody from the NIN song, possibly with some of the pathos of Johnny Cash as well.

You can get llama CPP or kobold.cpp binaries and load a quantized model right into them on the CPU only, no need to install Python or have an Nvidia GPU.

It might be better if you can spin up separate LLM's for each one, each with its own unique context or even fine tuning (or, at least, start each one's response with a fresh session using its own individually-tailored prompt and context).

One approach that I've been using with a local LLM, mostly brainstorming for my friends and my dungeons and dragons sessions, is to set up a prompt to be completed with some certain detail of storyline, character background, etc., Then running that prompt maybe 50 or 100 times and keeping informal statistics on the different results that come out. In such an application, it can be used as much for inspiration of how not to sound bland, commonplace, or derivative. You can pick the one in 100 oddity that strikes you as interesting, or simply make sure not to use any of the outcomes that the LLM came up with.

Isn't the majority of what Snowden leaked still being privately held from the public?

From the AP story on this:

> Senate Majority Leader Chuck Schumer, D-N.Y., has said he will introduce legislation to regulate AI. He has held a number of briefings with government officials to educate senators about an issue that’s attracted bipartisan interest.

Is Chuck Schumer considered to be a good educator about AI? Isn't this typically the domain of the Senate Intelligence Committee?

> Today when you will buy "connected" car the manufacturer wants YOU to pay the internet bill for your car.

My Toyota has an optional subscription to allow its data connection to be used for roadside assistance services, but the connection is always there regardless (and is used by Toyota to push updates to the car's touchscreen software without me paying for any subscription).

> it's specifically GenAI where authors can't prove or don't have rights to content.

Even more specifically, the author admitted to the images being "obviously AI generated" and Valve alleges that the images themselves in the game's initial submission contained copyrighted third-party content.