furyofantares

People have been saying LLMs have hit a wall for like 4 years. People still say it today. I don't think it's been true at any point.

Sure, but I imagined anthropic would do that too and they didn't. Imagining it doesn't help!

It kills me to run out of claude code usage then my much more mundane usage, queries or web searches, is also cut off. Not doing that has been a big plus of chatgpt/codex in my mind.

Remarkably the source is so small I can just post it here.

> Let's write our own YouTube clone based on whatever videos have been recommended on Hacker News.

As a game designer the answer is very obvious: That's hard work!

As a heavy LLM user there is another obvious answer: The LLM is very good at copying.

I suppose those are two sides of the same coin. All the hard decisions and all the creativity is already done.

Those aren't the only options though. I've long told people that want to get into game dev that the best thing to do is mod their favorite games. I'm hopeful that the technical ease to do that now leads to a lot of human creativity being expressed.

I did misspeak by saying production, but the context was google products and I mentioned size thrice.

Production was probably the wrong word - I really mean size and complexity of the project.

A replacement for a complex UI that would take a human developer 6 months to do is indeed something I would consider a small project at this point.

They simply aren't as useful for production stuff as they are for little projects.

They are useful! Just not the 1000x return you get on a tiny thing or the 100x return you get on a small thing.

The problem is sharing it to a group of people who mostly aren't paying for it. It's a good way to get a thread that's just people reacting to the title, which drives HN toward just being a discussion board instead of a place to find good content.

I don't trust LLMs to give me numbers like this, and it's not just that it might make a mistake on the number or it might have garbage in its context - it also might just not understand the mildest nuance in its context or apply any common sense.

I recently asked what percentage of ice cream is air. It told me up to 100%. When I followed the sources, as I always do and not just with silly answers like that, the number in its context was indeed up to 100%. There's an industry term called overrun, measured in percentages. But 100% overrun means it's 50% air.

So why would I trust any of these numbers in this obviously LLM-generated and probably LLM-researched article?

I guess it is slightly clever, if you feel like slipping out posts, to make them anti-AI. Really blunts the effectiveness of criticism of putting slop out when the slop agrees directionally with the criticism.

It's extremely inaccurate to say 80% of the gamblers were convinced because that's not how it works at all. Each bet has someone on the other side of it - 50% of the people bet on Bolsanaro and 50% against.

If you bet on 20% chances and they happen 30% of the time you make money. The payoffs are such that you win by being a better predictor than the market. I'm fine criticizing it as a gambling blight (not to mention a corruption payoff) but it is also a prediction market.

20% happens about 1 in 5 times. I know that's snarky wording but 20% shots come though constantly in daily life.

I think as humans we are really bad at intuition about probabilities, and it being an election only makes it much worse. I don't think anyone is consciously conflating these with poll numbers but percentages of voters are the numbers we see a lot more than prediction markets, and 80% is idunno an 80% chance of getting between 50% and 54% of the vote or something.

The only other post on this blog is "Anything connected to the internet is going to get hacked to pieces." I'm not claiming any contradiction but it is funny.

I think there is a possible future where "I" am a talking cucumber, though?

Ignoring that you have to give up a sense of personal identity at this point, there is nothing in the laws of physics that says the matter that makes up what we're presently calling me can't morph into a talking cucumber with a brain with my memories in it. It's just absurd number of extremely specific things that have to happen out of an ever more absurd number of possible things to happen.

For most things I find medium to be the highest quality output. However if I want it to do spatial reasoning want it be funny / write something (not code) of almost-passable quality, then xhigh is where it's at.

Yeah.

System 2 is slow because it is sequential. Every output depends on all the prior outputs. Like a transformer. (Even if this comes about as the result of massively parallel processes, both in the brain and in the model.)

System 1 is more of a parallel thing, operating on the output as a whole rather than in sequence. Much more like diffusion.

I wonder if you could exfiltrate some weights, and gt them to do inference.

I would guess Linus has reviewed more code than anyone else in the world, so in that sense I'm not that surprised.

I also don't think the C++ thing was out of a stubbornness for the old way of doing things, but had concrete reasons to reject it. I guess it was a little of both maybe 20 years ago.

But there's Rust in the kernel now too.

I knew the middle of the alphabet wouldn't be the correct cut but figured people would get the gist.

If you're putting a book out there of predictions it's probably trying to inform people of something they don't see coming, something they would have guessed much less than 50% while you think you have special knowledge to impart that it is indeed more likely to happen than they think.

Nobody would buy a book predicting the price of bitcoin in 10 years will end in an even number, or that the president's name will start with a letter in the first half of the alphabet.

But if someone says here's some wild sounding stuff that I have good reason to believe is coming, that's more interesting.

I used a game framework I made for myself that I use for a dozen different projects. It's written in Rust with all game logic written in AssemblyScript. I also have a library for managing state in a way that forces game rules to be completely divorced from UI, which I think keeps LLMs on the rails. It makes everything highly testable and prevents a certain type of ball of mud architecture that vibed apps tend to in my experience.

I should have mentioned that I think the bots are decent for the games I personally play - but they are probably bad for other games. In some cases the bots are purely heuristic based but in others they are informed by a small trained neural network or two. Hearts bots, for example, have two networks. One predicts what cards are in other players' hands, which is an easy network to train, and the other ingests that info alongside game state and chooses what to do, and was trained on self-play.

These models will be interacting from with random websites at scale once they're released into claude code.

Makes the victim more real (to the judge), I guess. And the family are victims too, the ones who continue to suffer, so I guess the 4D chess is demonstrating what the victims lost.

It eventually makes sense to my brain, aside from the AI part of course. If I'm a remarkably forgiving person my family might both appreciate that about me, and not be remarkably forgiving themselves. And they might say look, we lost this remarkable human being, punish the responsible party harshly.

That this is just model capabilities and not swarms of agents is dizzying to me. How long until we get access to these capabilities? How long until we can run something like it locally?

And what the hell will the frontier labs have by then?

Maybe I'm overreacting, I'll have to screw my head back on before I can process this.

Yeah well I'm doing everything in AssemblyScript.

That's cool.

I have agents going pretty much nonstop for hobby projects. This is great value for me because I get to see a lot of things I'm interested in come to life. It's very token-inefficient though and if usage limits were reduced much this wouldn't be worth doing.

For my actual job though, I don't let agents go for hours on end and I care about the code. I spend between $1000 and $2000 per month there. Of course we pay API pricing. I wonder how much you're getting for that $20 in API pricing, it could be 10x or even 50x. At least on the max plans it can be that high if you are hitting usage limits, I don't know about the 20.

We will use our big brains to draw bold conclusions about the most effective use of money, without the burden of conventional wisdom.

For our first bold conclusion we have determined that all of the world's money should be in the hands of people with big brains who intend to draw bold conclusions about the most effective use of money, without the burden of conventional wisdom.

This concludes the planning phase. There will be an additional planning phase once we've collected all of the money, we promise.

> About Me

> At my core, I am a creative. Whether it's games, videos, code, or music, I'm at my best when I'm creating something.

They've just handed you power tools.

There is a sadness in learning that I am less creative than I thought, that much of my self-identifying as creative came from having mastered some technical skills that used to be required to express my creativity.

Now people without those skills can express their creativity in the same domains. It's a great benefit for the world, I expect, but takes a minute to get used to the leveling of the playing field in areas I'd spent time mastering.

I won't try to convince anyone with specifics but it was definitely LLM generated. I think by someone who knows what topics they want to cover and did have something they wanted to post about. At a high level it is an infodump with almost no opinion expressed.

I learned a few things from it and have some interest in the topic. So this sort of a weird valley where I'm very annoyed by the writing but also getting something out of it. And it's more than I would have gotten out of promoting an LLM myself.

Which I know, because I have written a blog post about SDFs before, and I tried to author it with an LLM helping and the text just came out awful. I think I rewrote my entire post eventually - I am fairly confident at least that none of the LLM's words remained. I mention it because I've already read the SDF section of this blog largely, when an LLM generated mostly the same text 5 months ago.