smokel

17 ms·

271.
by smokel 1y ago
> CUDA isn't really used for new code. I don't think this is particularly correct, or at least worded a bit too strongly. For Nvidia hardware, CUDA just gives the best performance, and there are many optimized libraries that
272.
Reinforcement Learning Conference 2025: Outstanding Paper Awards(rl-conference.cc)
1 points by smokel 1y ago|0 comments
273.
by smokel 1y ago
The 20B version doesn't fit in 10GB. That might explain some issues?
274.
by smokel 1y ago
That was before the invention of fire? I think we had even worse problems back then.
275.
by smokel 1y ago
The premise of this article seems to be that the researchers at Windsurf have been able to learn amazing things thanks to near unlimited VC money, increasing their value to astounding numbers. That may be true (although I doubt the numbers
276.
by smokel 1y ago
> The underlying technology is pretty lame. This depends on the perspective. Take a step back and consider what the actual technology is that makes this possible: neural networks, the transistor, electricity, working together in group
277.
by smokel 1y ago
It's slightly more complicated than that. If people work less, they make less money, and that means they can't buy a house, to name just one example. Housing is not getting any cheaper for a myriad of reasons. The same goes for
278.
by smokel 1y ago
> The company that creates an AGI first will win and get the most status. I doubt it. History has shown that credit for an invention often goes to the person or company with superior marketing skills, rather than to the original creat
279.
by smokel 1y ago
No. You might get it to run, but you would also get old security exploits to run.
280.
by smokel 1y ago
I don't think understanding leads to fun in most people. It probably is the case for most people here on Hacker News, but I doubt it's universal.
281.
by smokel 1y ago
Perhaps you are thinking of Daniel Rozin's "Wooden Mirror" (1999)? https://www.smoothware.com/danny/woodenmirror.html
282.
by smokel 1y ago
I was going for irony, not analogy. Unfortunately, even though some incompetent fools think it is, life is not a game.
283.
by smokel 1y ago
It's similar to how, near the end of a Monopoly game, a player might indiscriminately hand over a stash of $100 bills to acquire Mediterranean Avenue, even though the property is mortgaged.
284.
by smokel 1y ago
Or it might go up.
285.
by smokel 1y ago
If someone is hired, they will most likely continue for a long time. They will make friends, spend time learning the domain, and a company will invest quite some hours. During the interview phase, it is easier to swap candidates.
286.
by smokel 1y ago
Unfortunately it is not that easy to simulate traffic, especially not on city scale or larger. The most important input to a traffic simulator such as this is the so-called "traffic demand", i.e. the routes that vehicles follow.
287.
by smokel 1y ago
They are easier to make than full-fledged user interfaces, so you get to see more of them.
288.
by smokel 1y ago
Have you tried that with the code repository that we're discussing here? It took this trained professional over an hour to get started, and then I gave up. It would take an additional 24 hours and quite some hardware simply to reprodu
289.
by smokel 1y ago
One aspect I find puzzling is why most two-factor authentication (2FA) applications restrict authentication to only a single valid code at any given time. This constraint inevitably creates a window during which it is inconvenient or imprac
290.
Ask HN: Why the fundamental skepticism around LLMs?
3 points by smokel 1y ago|15 comments
291.
by smokel 1y ago
Most LLM user interfaces, such as ChatGPT, do have a memory. See Settings, Personalization, Manage Memories .
292.
by smokel 1y ago
You touch upon an interesting phenomenon that is quite pervasive throughout the world of software development, and possibly many other fields as well. To avoid heated discussions, allow me to illustrate the concept with why enterprise softw
293.
by smokel 1y ago
I sympathize partially with your views, but how would this work in practice? Where would the review comments be stored? Is one supposed to browse Hacker News to check the validity of a paper? If a professional reviewer spots a serious pro
294.
by smokel 1y ago
Hm, I'm not so sure it's fair play for the Sudoku puzzle. Suggesting that the AI will understand the rules of the game with only 1,000 examples, and then adding 1,000,000 derived examples does not feel fair to me. Those extra ex
295.
by smokel 1y ago
To follow up, after experimenting a bit with the source code: 1. Please, for the love of God, and for scientific reproducibility, specify library versions explicitly, and use pyproject.toml instead of an incomplete requirements.txt. 2. The
296.
by smokel 1y ago
Looking at the code, there is a lot of data augmentation going on there. For the Sudoku and ARC data sets, they augment every example by a factor of 1,000. https://github.com/sapientinc/HRM/blob/main/dat
297.
by smokel 1y ago
Thanks for pointing that out. To be fair, MuZero only learns a model of the rules for navigating its search tree. To make actual moves, it gets a list of valid actions from the game engine, so at that level it does not learn the rules of t
298.
by smokel 1y ago
If I understand this correctly, it learns the rules of Sudoku by looking at 1,000 examples of (puzzle, solution) pairs. It is then able to solve previously unseen puzzles with 55% accuracy. If given millions of examples, it becomes almo
299.
by smokel 1y ago
Any practical tips on how to improve upon the current situation?
300.
by smokel 1y ago
There are many systems written in Cobol, Delphi, C++, that would really benefit from not being written in those languages. There are serious headaches to be had when you have to maintain software for ten years and up. Porting systems to ano

More ›