nl

35 ms·

751.
by nl 9mo ago
Not exactly sure what you are looking for here. That GRPO works? > Group Relative Policy Optimization (GRPO), a variant reinforcement learning (RL) algorithm of Proximal Policy Optimization (PPO) (Schulman et al., 2017). GRPO foregoes th
752.
by nl 9mo ago
wow I love it.
753.
by nl 9mo ago
> You have access to the best models & yet you still haven't managed to prompt it to give you the code to prove your point so spare me any further theoretical responses. Either show the code to do exactly what you're saying
754.
by nl 9mo ago
Sure. The point is that your statement about the ability to do RL is wrong. Additionally your response to the Deepseek paper in the other subthread shows profound and deliberate ignorance.
755.
by nl 9mo ago
> Scamming of gambling addicts is tragic but not detrimental to society. This isn't true. each 10 per cent increase in gambling expenditure in NSW results in more than 4,500 additional assaults 2,800 additional home bre
756.
by nl 9mo ago
Post RL you won't need to put the grammar in the prompt anymore.
757.
by nl 9mo ago
Disagree about Codex - it's great at doing things too! Gemini either does a Monet or demolishes your bathroom and builds a new tuna fishing boat there instead, and it is completely random which one you get. It's a great model but
758.
by nl 9mo ago
I guess the op was implying that is something fixable fairly easily? (Which is true - it's easy to prompt your LLM with the language grammar, have it generate code and then RL on that) Easy in the sense of "it is only having enoug
759.
by nl 9mo ago
I have a vibe coded fantasy console. Getting Doom running on it was easy. Getting the Doom sound working on it involved me setting there typing "No I can't hear anything" over and over until it magically worked... Maybe I s
760.
by nl 9mo ago
> My understanding/experience is that LLM performance in a language scales with how well the language is represented in the training data. This isn't really true. LLMs understand grammars really really well. If you have a gramm
761.
by nl 9mo ago
One of the interesting things to me about this is that Codex 5.2 found the most complex of the exploits. The reflects my experience too. Opus 4.5 is my everyday driver - I like using it. But Codex 5.2 with Extra High thinking is just a bit
762.
by nl 9mo ago
It's called Gas Town for a reason...
763.
by nl 9mo ago
The irony is that I submitted one of my open source projects because it was vibe-coded and people accused me of not vibe coding it !
764.
In Defense of Data Centers(deeplearning.ai)
1 points by nl 9mo ago|1 comments
765.
by nl 9mo ago
I think Andrew Ng wrote a great piece on this. For example, in the US, which do you think uses more water: Golf Courses or Data Centers? a) Gold Courses use twice as much water as Data Centers b) About the same c) Data Centers use t
766.
by nl 9mo ago
There's an update from Tao after emailing Tenenbaum (the paper author) about this: > He speculated that "the formulation [of the problem] has been altered in some way".... [snip] > More broadly, I think what has happen
767.
by nl 9mo ago
Isn't this just an awkward way of adding an extra layer to the NN, except without end-to-end training? Models like Stable Diffusion sort of do a similar thing using Clip embeddings. It works, and it's an easy way to benefit from t
768.
by nl 9mo ago
I think that just hasn't been updated.
769.
by nl 9mo ago
Interesting that in Terrance Tao's words: "though the new proof is still rather different from the literature proof)" And even odder that the proof was by Erdos himself and yet he listed it as an open problem!
770.
Erdos 281 solved with ChatGPT 5.2 Pro(twitter.com)
308 points by nl 9mo ago|294 comments
771.
by nl 9mo ago
This really needs to be `uvx` runnable.
772.
by nl 10mo ago
I got a fanless Celeron N4020 4GB RAM 64G Storage new on Amazon for under $150 in Jan 2025, and it has been running home assistant ever since. I don't think I could a RPi as cheaply once parts and power supply etc are taken into accoun
773.
by nl 10mo ago
Gemini 3 Flash is very good at the search task (it benchmarks quite close to 3 Pro in coding tasks but is much faster). I believe Amp switch to Gemini Flash for their search agent because it is better.
774.
Microsoft's spending on Anthropic AI on track to reach $500M(msn.com)
4 points by nl 10mo ago|1 comments
775.
by nl 10mo ago
Probably. Companies sell on efficiency when they know they lose on performance.
776.
by nl 10mo ago
> OpenAI has $1.15T in spend commitments over the next 10 years Yes, but those aren't contracted commitments, and we know some of them are equity swaps. For example "Microsoft ($250B Azure commitment)" from the footnote is
777.
by nl 10mo ago
> we demonstrated running gpt-oss-120b on two RNGD chips [snip] at 5.8 ms per output token That's 86 token/second/chip By comparison, a H100 will do 2390 token/second/GPU Am I comparing the wrong things somehow?
778.
by nl 10mo ago
So inference only and slower than B200s? Maybe they are cheap.
779.
by nl 10mo ago
> I am of the opinion that Nvidia's hit the wall with their current architecture Based on what? Their measured performance on things people care about keep going up, and their software stack keeps getting better and unlocking more p
780.
by nl 10mo ago
Why not ban computer games then?

More ›