Bjorkbat

It really is crazy how much was lost when Apple killed Flash. Absolutely miss Newgrounds. It's still around of course, I'm reflecting more on the vibes when it was in its heyday. Unbelievable the games people were making with Flash back then and how it spawned the careers of a ton of indie darlings. Also, not Flash at all, but does anyone remember Exit Mundi? Absolute gold.

Honestly, I kind of look back on blogging unfavorably. Before that people made websites to showcase their interests and hobbies, and because of that even the most basic looking websites could have a lot of "color" to them. Then blogging became a thing and people's websites became bland and minimalist. Arguably blogging culture is as responsible for the death of creativity on the internet as much as the constraints of mobile-friendly web design and Apple's aforementioned killing of Flash.

My thoughts as I reconcile my conflicting feelings on this. I think it should be objectively cool that Meta has finally managed to come out with a pair of smart glasses that come incredibly close to being a practical wearable.

The thing is, it's honestly hard to imagine doing anything cool with them. I think this has less to do with hardware limitations and more to do with vendor restrictions.

I think Meta is fundamentally incapable of making anything cool. Hence why they had to partner with Ray-Ban to make these glasses rather than making their own. I think Meta's failure to realize their version of the metaverse had to do with their inability to recognize coolness and taste as much as anything else. I think any and all apps Meta ships with these glasses are cursed to be a mediocre experience.

I think Apple could do a better job but at the end of the day I think the most interesting (not necessarily best) would be ones with the most developer freedom.

My personal favorite from that time was a website builder called "The Grid" which really overhyped on its promises.

It never had a public product, but people in the private beta mentioned that they did have a product, just that it wasn't particularly good. It took forever to make websites, they were often overly formulaic, the code was terrible, etc etc.

10 years later and some of those complaints still ring true

Ooh, fun, a FizzBuzzFeed quiz!

Sure, but if you're trying to get there by training a model on video games then you're likely going to wind up inadvertently creating a video game simulator rather than a physics simulator.

I don't doubt they're trying to create a world simulator model, I just think they're inadvertently creating a video game simulator model.

Also, to drive my point further home, in one of the demos they were operating a jetski during a festival. If the jetski bumps into a small Chinese lantern, it will move the lantern. Impressive. However, when the jetski bumped into some sort of floating structure the structure itself was completely unaffected while the jetski simply stopped moving.

This is a pretty clear example of video game physics at work. In the real world, both the jetski and floating structure would be much more affected by a collision, but in the context of video game physics such an interaction makes sense.

So yeah, it's a video game simulator, not a world simulator.

Genuinely technically impressive, but I have a weird issue with calling these world simulator models. To me, they're video game simulator models.

I've only ever seen demos of these models where things happen from a first-person or 3rd-person perspective, often in the sort of context where you are controlling some sort of playable avatar. I've never seen a demo where they prompted a model to simulate a forest ecology and it simulated the complex interplay of life.

Hence, it feels like a video game simulator, or put another way, a simulator of a simulator of a world model.

Somewhat related, but I’ve been feeling as of late what can best be described as “benchmark fatigue”.

The latest models can score something like 70% on SWE-bench verified and yet it’s difficult to say what tangible impact this has on actual software development. Likewise, they absolutely crush humans at sport programming but are unreliable software engineers on their own.

What does it really mean that an LLM got gold on this year’s IMO? What if it means pretty much nothing at all besides the simple fact that this LLM is very, very good at IMO style problems?

Alright, it does look pretty charming, and I especially like that it's open-source since pretty much anyone buying a domestic robot is likely to be a tinkerer of some sort, but at the same time it reminds me of the Jibo (https://robotsguide.com/robots/jibo).

For those who don't remember (I couldn't remember the name, only the face, had to look hard for it) it was a desktop robot released in 2014 that was hyped pretty hard at the time. It didn't help that the company that launched it was founded by a fairly well-known MIT professor.

And yeah, it was a flop. The $900 price tag wasn't helping things, but neither was the fact that it didn't really do anything that an Alexa couldn't. You bought it solely because you really liked the idea of robots and thought it was cool, not at all for its value around the house.

I'm not gonna dunk on this too hard since it's probably just a fun company side-project, but I might change my tune if they get too high on hype.

Yeah, this is something I find kind of tricky. I definitely believe that AI companies should get permission from rightsholders to train on their works, but actually compensating them for their works seems pointless. To make the royalties worthwhile you'd have to raise the cost per query to an absolutely absurd level

I recall it as less an evolution and more a complete tonal shift the moment o3 was evaluated on ARC-AGI. I remember on Twitter Sam made some dumb post suggesting they had beaten the benchmark internally and Francois calling him out on his vagueposting. Soon as they publicly released the scores, it was like he was all-in on reasoning.

Which I have to admit I was kind of disappointed by.

Personally I think a more effective analogy would be if someone used a textbook and created an online course / curriculum effective enough that colleges stop recommending the purchase of said textbook. It's honestly pretty difficult to imagine a movie having a meaningful impact on the sale of textbooks since they're required for high school / college courses.

So here's the thing, I don't think a textbook author going against a purveyor of online courseware has much of a chance, nor do I think it should have much of a chance, because it probably lacks meaningful proof that their works made a contribution to the creation of the courseware. Would I feel differently if the textbook author could prove in court that a substantial amount of their material contributed to the creation of the courseware, and when I say "prove" I mean they had receipts to prove it? I think that's where things get murky. If you can actually prove that your works made a meaningful contribution to the thing that you're competing against, then maybe you have a point. The tricky part is defining meaningful. An individual author doesn't make a meaningful contribution to the training of an LLM, but a large number of popular and/or prolific numbers can.

You bring up a good point, interpretation of fair use is difficult, but at the end of the day I really don't think we should abolish copyright and IP altogether. I think it's a good thing that creative professionals have some security in knowing that they have legal protections against having to "compete against themselves"

Something missed in arguments such as these is that in measuring fair use there's a consideration of impact on the potential market for a rightsholder's present and future works. In other words, can it be proven that what you are doing is meaningfully depriving the author of future income.

Now, in theory, you learning from an author's works and competing with them in the same market could meaningfully deprive them of income, but it's a very difficult argument to prove.

On the other hand, with AI companies it's an easier argument to make. If Anthropic trained on all of your books (which is somewhat likely if you're a fairly popular author) and you saw a substantial loss of income after the release of one of their better models (presumably because people are just using the LLM to write their own stories rather than buy your stuff), then it's a little bit easier to connect the dots. A company used your works to build a machine that competes with you, which arguably violates the fair use principle.

Gets to the very principle of copyright, which is that you shouldn't have to compete against "yourself" because someone copied you.

Honestly feels like the whole Soham Parekh thing on Twitter is one giant joke with the one sincere / honest remark being the original from @Suhail.

Like, I can't wrap my head around this many people having some kind of experience with a single guy who's claim to be fame is basically gaming the interview process at an incredible amount of Y Combinator startups.

Yeah, I was about to say, it sounds a lot like this guy is just riding an intense high from getting Claude to build some side-project he's been putting off, which I feel is like 90% of all cases where someone writes a post like this.

But then I never really hear any update on whether the high is still there or if it's tapered off and now they're hitting reality.

> These are just good chess moves. The "super-intelligence" bit is just hype/spin for the journalists and layperson investors.

Which is kind of what I figured, but I was curious if anyone disagreed.

It's a smart purchase, it's just that I don't see how these datasets factor into super-intelligence. I don't think you can create a super-intelligent AI with more human data, even if it's high-quality data from paid human contributors.

Unless we watered-down the definition of super-intelligent AI. To me, super-intelligence means an AI that has an intelligence that dwarfs anything theoretically possible from a human mind. Borderline God-like. I've noticed that some people have referred to super-intelligent AI as simply AI that's about as intelligent as Albert Einstein in effectively all domains. In the latter case, maybe you could get there with a lot of very, very good data, but it's also still a leap of imagination for me.

Fair enough. If you aren't willing to give your friend $14 billion to join your company so you can hang out more, then are you two really friends?

I don't actually think this is the case, but nonetheless I think it would be kind of funny if LLMs somehow "discovered" linguistic relativity (https://en.wikipedia.org/wiki/Linguistic_relativity).

Point remains though, they crushed the benchmark using a specialized model that you’ll probably never have access to, whether personally or through a company.

They inflated expectations and then released to the public a model that underperforms

Related, when o3 finally came out ARC-AGI updated their graph because it didn’t perform nearly as well as the version of o3 that “beat” the benchmark.

https://arcprize.org/blog/analyzing-o3-with-arc-agi

He's still a great designer, the problem though is that without the right kind of editorializing force he'll make mistakes, usually in the form of compromising practicality and functionality for the sake of aesthetics. I should probably clarify that this isn't some fault unique to Jony Ive. It feels like it's common among designers and probably just creative people in general.

I was about to comment on a past remark from Anthropic that the whole reason for the convoluted naming scheme was because they wanted to wait until they had a model worth of the "Claude 4" title.

But because of all the incremental improvements since then, the irony is that this merely feels like an incremental improvement. It obviously is a huge leap when you consider that the best Claude 3 ever got on SWE-verified was just under 20% (combined with SWE-agent), but compared to Claude 3.7 it doesn't feel like that big of a deal, at least when it comes to SWE-bench results.

Is it worthy? Sure, why not, compared to the original Claude 3 at any rate, but this habit of incremental improvement means that a major new release feels kind of ordinary.

A while back someone here on Hacker News made a pretty insightful comment that as great of a designer as Jony Ive is, a large part of his success is owed to the fact that he had an "editor" in the form of Steve Jobs. Once Jobs passed, he no longer really had an editor.

It remains to be seen whether Sam Altman / OpenAI in general will be a good editor

My weird theory: the alpha releases are more expensive than people realize, and OpenAI can afford to launch money-losing alphas on Pro because few users have Pro accounts and the lose less money per Pro user.

If anyone could use Codex, you’d want to price the tokens at a rate where you don’t lose money, or at least not too much money. If you price the tokens too high, then not only are people not using Codex, but customers and investors alike begin to lose faith, they no longer “feel the AGI”.

I kind of had the feeling LLMs would be better at Python vs other languages, but wow, the difference on Multi SWE is pretty crazy.

Fair point, we use metaphor to explain and understand a variety of topics, and a lot of those metaphors are best understood through pop culture analogies.

A reasonable compromise then is that you can train an AI on Wikipedia, more-or-less. An AI trained this way will have a robust understanding of Superman, enough that it can communicate through metaphor, but it won't have the training data necessary to create a ton of infringing content about Superman (well, it won't be able to create good infringing content anyway. It'll probably have access to a lot of plot summaries but nothing that would help it make a particularly interesting Superman comic or video).

To me it seems like encyclopedias use copyrighted pop culture in a way that constitutes fair use, and so training on them seems fine as long as they consent to it.

I broadly agree in that sure, unfettered access to copyrighted material will AI more capable, but more capable of what exactly?

For national security reasons I'm perfectly fine with giving LLMs unfettered access to various academic publications, scientific and technical information, that sort of thing. I'm a little more on the fence about proprietary code, but I have a hard time believing there isn't enough code out there already for LLMs to ingest.

Otherwise though, what is an LLM with unfettered access to copyrighted material better at vs one that merely has unfettered access to scientific / technical information + licensed copyrighted material? I would suppose that besides maybe being a more creative writer, the other LLM is far more capable of reproducing copyrighted works.

In effect, the other LLM is a more capable plagiarism machine compared to the other, and not necessarily more intelligent, and otherwise doesn't really add any more value. What do we have to gain from condoning it?

I think the argument I'm making is a little easier to see in the case of image and video models. The model that has unfettered access to copyrighted material is more capable, sure, but more capable of what? Capable of making images? Capable of reproducing Mario and Luigi in an infinite number of funny scenarios? What do we have to gain from that? What reason do we have for not banning such models outright? Not like we're really missing out on any critical security or economic advantages here.

I mean, I don't disagree with you when you say that something that would take an hour or more to implement would only take 10 minutes or so with AI. That kind of aligns with my personal experience. If something takes an hour, it's probably something that the LLM can do, and I probably should have the LLM do it unless I see some value in doing it myself for knowledge retention or whatever.

But working on features that can fit within a timebox of "an hour or more" takes up very little of my time.

That's what I mean, there are certain contexts where it makes sense to say "yeah, AI made me 2x-10x more productive", but taken as a whole just how productive have you become? Actually being 2x productive as a whole would have a profound impact.

I'm always a little bit skeptical whenever people say that AI has resulted in anything more than a personal 50% increase in productivity.

Like, just stop and think about it for a second. You're saying that AI has doubled your productivity. So, you're actually getting twice as much done as you were before? Can you back this up with metrics?

I can believe AI can make you waaaaaaay more productive in selective tasks, like writing test conditions, making quick disposable prototypes, etc, but as a whole saying you get twice as much done as you did before is a huge claim.

It seems more likely that people feel more productive than they did before, which is why you have this discrepancy between people saying they're 2x-10x more productive vs workplace studies where the productivity gain is around 25% on the high end.