hodgehog11

7 ms·

91.
by hodgehog11 4mo ago
I think we're starting to see more of this sort of thing happening now with Proton and Wine gaining prominence in the Linux community. Some games (Elden Ring comes to mind) have bad enough PC ports when they come out that the compatibi
92.
by hodgehog11 4mo ago
I was mostly referring to Volume 1 (not advanced topics). You have a point that Volume 2 definitely contains more. To be honest, I was mostly covering myself from a "that's not real deep learning" criticism; "relatively
93.
by hodgehog11 4mo ago
Books? No, not really. Maybe others will have better suggestions for newcomers, sorry. Are you talking research novelty or just applying current methods to a given task? The latter is covered well by Andrej Karpathy's videos and by jus
94.
by hodgehog11 4mo ago
Obviously it's great that those who are only aware of JEPA should be educated about CCA. If you don't know CCA, you should not be working in unsupervised learning. However, it's pretty obvious that they are related since CCA
95.
by hodgehog11 4mo ago
Yes I have tried giving these same prompts to other models. The difference has been painfully clear. Switching back to Opus, it is completely unable to do anything that I had asked of Fable without significant conceptual and engineering err
96.
by hodgehog11 4mo ago
> You act like we all should be entitled to these completely unnecessary things. Living standards should not be worse for younger generations than older generations. But more importantly, younger generations spend money on ridiculous &qu
97.
by hodgehog11 4mo ago
Interesting, thanks for sharing. That is something I would have expected it to do well on, unless it tripped the internal rerouting. My experience on computational geometry problems has been universally positive (virtually flawless), and fa
98.
by hodgehog11 4mo ago
> Huh? What luxuries? Us plebes can't fly business class? We can't buy that expensive handbag? Yes. Why do you need to ask this? This is the K-shaped economy. Demand is dropping (especially in luxury handbags) as the middle cla
99.
by hodgehog11 4mo ago
> Fable is a general purpose model for use in regular chat, analysis, as well as coding This is a forum filled with experts. Putting marketing aside, in a forum like this, it is most useful to assess models according to the toughest prob
100.
by hodgehog11 4mo ago
Some very tough computational geometry problems I couldn't solve on my own, nor with the assistance of other AIs or my colleagues. Fable did them all first go. The most impressive built a custom optimizer with a ludicrous number of ada
101.
by hodgehog11 4mo ago
It's really annoying when people seem to assume that their experiences and use-cases are universal, and that other experiences are "psychosis". Fable seems to excel at tasks that are research-level hard. The vast majority of
102.
by hodgehog11 4mo ago
Same experience. Wouldn't waste my tokens on easy stuff for it. It blasted through some of my toughest problems and produced some truly great code.
103.
by hodgehog11 4mo ago
This is a ridiculous thing to test on it. Other models are trained on that kind of thing, use those instead. Fable was designed for _really_ hard software engineering problems. Possibly large, but especially hard. For those tasks, you feel
104.
by hodgehog11 4mo ago
This is utterly daft to say if you actually used the thing for hard problems, something that benchmarks have been known to be unable to capture. It is night and day compared to Opus and every other model out there. It was nice while it last
105.
by hodgehog11 4mo ago
> No actual facts in your argument I don't see facts in any of these arguments, that's really the point. Doomerism isn't particularly productive, but I'm tired of the complacency and the suggestions that everything wi
106.
by hodgehog11 4mo ago
I generally agree with this rebuttal. Each KAN layer is more expressive on a per-layer basis, although there is a mapping to an MLP with more layers. With the current hardware implementations, yes, MLPs have an advantage overall. I can cert
107.
by hodgehog11 4mo ago
The benefit in KANs is interpretability, not expressivity. It's a structure that lends itself well to performing symbolic regression or other interpretable downstream tasks. This can make it better suited for scientific tasks, for exam
108.
by hodgehog11 4mo ago
Others have commented on the rate of AI improvement. It doesn't need to be current rate for it to be an even more serious problem in the very near future. That's irrespective of prior booms. Regarding AI companies having capital t
109.
by hodgehog11 4mo ago
I get about the same success rate with my problems (scientific computing usually), but they're often _much_ easier to check than to write, so an 80% success rate becomes game-changing.
110.
by hodgehog11 5mo ago
I appreciate the diplomatic approach. Disagreements about the nature of consciousness are perfectly natural in this context, and concepts such as "understanding" or "cognition" are often difficult to effectively define.
111.
by hodgehog11 5mo ago
From Merriam-Webster: cognitive: as in reasonable; of, relating to, or involving conscious mental activities (such as thinking, *understanding*, learning, and remembering)
112.
by hodgehog11 5mo ago
"LLMs could not possibly be good models of some cognitive capacity because they are just a bunch of numbers guessing the next word. They have no linguistic module, so they cannot exhibit cognition". That's the argument. It&#x
113.
by hodgehog11 5mo ago
I supervise quite a few Masters students. In my particular setting, believe me, LLM stupid for the top three chatbots is easier to work with than real human stupid now. We passed that threshold earlier this year.
114.
by hodgehog11 5mo ago
As an example, "They're made out of weights" describes why the weight-based construction of neural networks should impact the way that you think about them and their outputs. I would argue that an offhand description of its m
115.
by hodgehog11 5mo ago
Are you serious? I hear it every single day, especially from computer scientists. There are top ranked posts here on HN _today_ with this argument.
116.
by hodgehog11 5mo ago
This is the whole spirit of the "Stop Killing Games" argument though: you don't need to keep any of that stuff once support drops. It just needs to be "functional", in the most basic possible sense. If there are no
117.
by hodgehog11 5mo ago
No offense intended, but this sounds like you're projecting impressions about LLM output for programming. Your words make a lot of sense to me in that context, but not as much here. I can speak with a reasonable amount of experience he
118.
by hodgehog11 5mo ago
ChatGPT and Gemini are actually fairly comparable. Claude has been utterly useless with most math problems in my experience because, much like less capable students, it tends to get overly bogged down in tedious details before it gets to th
119.
by hodgehog11 5mo ago
This is beyond ridiculous to say considering whose blog this is. For those that don't know, this is Timothy Gowers. He is one of the most accomplished mathematicians in the world. Like Terence Tao, he is considered one of the world lea
120.
by hodgehog11 5mo ago
May I remind you that this is Timothy Gowers. He says he doesn't understand, but he most certainly has far greater capacity than most to detect complete junk from a maybe plausible argument. His colleague is even better able to judge t

More ›