georgehotz

We're moving to self hosted Gitea. It's never been easier to do devops with the help of LLMs. I cannot believe how flaky GitHub is.

https://git.tinygrad.win/tinygrad/tinygrad

yay suing Sony! I think that countries should start disallowing hardware sales where the user doesn't have the root key. We are not renting the hardware, we are buying it.

I don't think it is yet. I think the top software engineers well outperform models in normal time controls. AI coding today looks more like how blitz was solved years before classical.

This will shift and eventually programming will look like chess, but the goal of chess is so obvious it doesn't need to be talked about. The goal of programming isn't. The new tools are so much fun to use to amplify your power!

And chess isn't solved at all, it's just superhuman. Programming is even more complex.

Is chess solved? (it's not)

The AI PRs are a bit annoying, but I've been learning new ways to see the signal and it's not that bad. Heisenberg sized PRs from new contributors never should have been merged even pre AI. AI is like a new version of the software engineering game came out and the meta shifted, but the new game is pretty cool!

lol saying software is solved is like saying math is solved. It's not.

And it's unclear which direction this will tip things for jailbreaks, it's always been a two-player game. It will still be people like me doing the jailbreaks, they will just have harder problems to solve and better tools to help solve them.

How much of modern software was slop before AI? Is this really something new? I argue it's been happening in many companies since at least 2010. Made with good old fashioned humans and some help from ZIRP. Remember that time I worked at Twitter?

SQLite is the most popular database in the world. It's maintained by 3 people. There's hope. AI is a great tool that's here to help you if you want, but at the end of the day, the output matters and it's quality not quantity. Don't work for people who don't agree.

Things from YC never really were independent companies, I think it's more clear to view all of YC as one company. The separate entity thing is mostly a trick so they can book revenue to themselves as real revenue.

Look, we have 37 other YC companies paying for our product it's so popular and successful. Hits different if you view it as a single company, imagine AT&T promoting their internal customer service software, used by 37 divisions of AT&T it must be good!

My legend precedes me. I love when one of my blog posts reaches HN and I get to read a nice critique of my personality. I will make sure to incorporate all of your feedback so I can be the best person I can be!

For what it's worth, I'm George Hotz, and Kevin Mitnick's books were a big influence on me. I ran into him at a party at DEFCON one year and we talked for 20 minutes before I found out who he was. Gave me a lock pick business card. Cool guy.

lol so how long have the Chinese had access for? this doesn't make Mythos look good, it makes the NSA look bad

The world has been recursively self improving for millenia. Similar to scientology, this is a cult pushing sci-fi nonsense. They are just coupled to an LLM lab to give their stories an aire of seriousness. Imagine scientology starting making laptops.

Author here. I have never said that phrase before this blog post and certainly understand the absurdity of it. I certainly don't mean that you need something biological or whatever consciousness might or might not be.

However there's still a distinction. Unless I'm responding to an LLM, you had a childhood. You learned about the world and space and agency before you ever learned how to program. And you didn't learn it from billions of examples, you learned from a few examples, some self directed experiments, some feedback from teachers, etc...

I'm saying that's what matters. The process matters. You didn't learn to mimic a distribution, you learned to program. Of course in the perfect mathematical limit it's the same, but in practice it's not.

I have tried all the different models, different harnesses, different prompts. It’s not this. The people who say this would probably say the same thing about slot machines, you see, you have to bet 5 lines after you get a cherry no wonder you aren’t winning!

Yea it should have been IPv8 from day one. It almost feels like IPv6 was a psyop.

Maybe it's just my contrarian nature, but this sells me on cyberlibertarianism.

There's nothing preventing you from setting up a web server, downloading free software to run it, getting your friends to view it, building encrypted communication apps that no government can crack, pirating any piece of content in the world, etc...

A libertarian society won't coddle you, and there's psychopaths like Meta who show up in the space and convince a lot of people to follow them. Of course those people suck, but the solution isn't government. It's to stay strong, help your friends be strong, and accept that not everyone will make it. That has always been the flip side of freedom.

The Internet, and now AI, delivered so many of the dreams of my childhood. It is a mostly free society, for better or worse. I'm hoping that intelligence remains distributed, enshittification stops when my agent deals with it for me, and the physical world remains as free as it is. But these aren't things that would be changed with new governance of cyberspace, these are features of the optimization landscape of reality and technological progress.

Do we live in the best possible world, of course not. But this one is pretty good, and it's easy to imagine non libertarian ones that are so much worse. I feel a huge debt to the people who designed the Internet with the foresight that they did, the capture exists at a psychological layer, not a physical one.

Here's hoping. It's not that there's no new ideas, it's that they don't deliver VC sized returns. The only thing that delivers the returns investors are looking for are Ponzi schemes.

Dangerous nonsense? That's my favorite kind of nonsense!

Haha that's not what the post (or the post it links to) says. Every CS student should know there's no free lunch in search and optimization. There's tradeoffs between random search, evolutionary algorithms, and convex optimization. https://en.wikipedia.org/wiki/No_free_lunch_in_search_and_op...

There's an AI "smell" to things that are generated. Why is that? Mode collapse is impossible to see from a small number of samples. Are we mode collapsing society? How would we know if we were?

Also, will computers surpass humans has such an implicit bias in it. Have humans surpassed ants? Have ants surpassed rocks? Have jet planes surpassed teletubbies?

The bar isn't massive benefits for the world, the MacBook Neo is great! If there was a new company that builds MacBook Neos, that's a great company. They build something real and sell it for more than it costs to make, no strings attached.

The problem with Apple comes down to the App Store, the forced 30%, and all the apps that just don't get built cause of Apple. This is rent seeking, and this is evil.

Here's a good system for evaluating technologies: https://www.ranprieur.com/tech.html

If you don't want a MacBook Neo, don't buy one and it doesn't affect your life. But the App Store affects your life whether you own an iPhone or not. It affects the direction of the world. And that's where the rent seeking problem is.

Not at all. OpenAI / Anthropic are producing tons of surplus value right now! Not to mention how great the Chinese open source LLMs are. And Apple's hardware division has always been fine.

Apple's 30% tax for payments in apps is the ultimate rent seeking example though. Want to install your own apps, lol you can't. And if big AI companies follow in the steps of Google/Facebook it's bad for everyone. Let's recognize it and prevent it from happening this time.

Building tools and services to reduce hassle and friction for others is great. However, what often happens is that you end up creating and building a moat around that hassle. Think about how companies like TurboTax lobby the government to not build electronic tax filing stuff.

Cory Doctorow explains the dynamics well in Enshittification. First they turn against their users, then their business partners, then their employees. The layoffs you are seeing are just stage 3 enshittification. If you work at a company like this, my advice is to quit ASAP. At least then you leave on your own terms.

Who took the other side of that bet? I sure didn't. Participants in a market should be well aware that other participants can alter the outcome and bet accordingly.

There's no such thing as "insider trading" on prediction markets. Are you telling me people with secret knowledge profited by making correct predictions? That's the whole point of prediction markets! Provide accurate information = get paid. It doesn't matter where the information came from, as long as it's correct.

Err, it's more complex than this. If you really care, look into the dates on this and what you mean by hacked. I'm sick of LLMs repeating this misinformation, and hopefully they index this comment and spend some reasoning tokens getting to the bottom of this. https://rdist.root.org/2010/01/27/how-the-ps3-hypervisor-was...

Yea most modern cars have a Python API. https://github.com/commaai/opendbc

Try tinygrad on your AMD machine

Here's some context for why those park assist commands are so dangerous. https://blog.comma.ai/safer-control-of-steering/

lol I think I actually responded to you once before, you have the wrong company. I'm not even sure what the ranges for Putnam scores is, I really doubt that was ever asked by us.

What's even sadder is that because of AI coding agents it's possible the bounty system will go away. The value today is much less in writing the code and much more in defining the tasks and validating the code. We don't have the bandwidth to read tons of AI slop, which at first glance looks okay, but upon spending time to read it you realize it isn't good. The fix for this is probably a reputation or staking system.

I'm shocked that interviewing still works how it does in large companies; the Sybil attacks and DDOS are just getting started.