Interesting that 3 names I recognized as physicists from stat mech adjacent fields. They continue to punch above their expectations (as sampled by general dismissal of physicists in AI/ML on HN and reddit).
Do people not like physicists?
NeurIPS 2025 Best Paper Awards
11–20 of 31 posts
Re: NeurIPS 2025 Best Paper Awards
#12From the figure in the first paper listed: > Responses to the query “Write a metaphor about time” clustered by applying PCA to reduce sentence embeddings to two dimensions. […] The responses form just two primary clusters: a dominant cluster on the left centered on the metaphor “time is a river,” and a smaller cluster on the right revolving around variations of “time is a weaver.” I just gave Gemini 3 the same prompt…
Constantly flowing and makes things smooth like river stones; compared to Tait's "time is a series if staric pictures", Gemini's output is not so different from a river metaphor.
Re: NeurIPS 2025 Best Paper Awards
#13Are there any talks about these papers on youtube or somewhere? I think I find it easier to listen and watch then read or maybe I'm just lazy, not sure.
most papers have slides with audio, and some, including the awards ones will have short frontal talks. this will be released at some point after the conference, but right now looks like you'd have to be registered to see it.
Re: NeurIPS 2025 Best Paper Awards
#14Are there any talks about these papers on youtube or somewhere? I think I find it easier to listen and watch then read or maybe I'm just lazy, not sure.
use NotebookLM
Re: NeurIPS 2025 Best Paper Awards
#15I think my favorite of the bunch is the "Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model" paper. Easy to read, gets the point across very intuitively and quickly, and the point is very interesting and relevant to a lot of people. About the Superposition paper - this is close to what I've been thinking about over the past week. I'm thinking that concepts or choices in a…
I am not sure how to interpret the first paper's results.
If we use a random number generator then we will converge to 100% correct answers under pass@n in the limit.
A random number generator will eventually outperform or match all models (for large n) whenever top-p is less than 1 because the other models will most likely have some level of bias that makes correct CoTs mathematically impossible due to the tokens being too improbable and being filtered out by top-p, meaning that other models will asymptote to below 100% while the RNG will reach 100% in an almost surely sense.
Under this paper's logic doesn't that mean that the random number generator is a superior reasoner?
Re: NeurIPS 2025 Best Paper Awards
#16Are there any talks about these papers on youtube or somewhere? I think I find it easier to listen and watch then read or maybe I'm just lazy, not sure.
use NotebookLM
Whenever I search for the title a new machine learning paper, there are a bunch of YouTube videos about it which are just NotebookLM slop. It's straight up environmental pollution.
Re: NeurIPS 2025 Best Paper Awards
#17Oh man, this link is worth it just for the "Reflections from the Selection Committee."
These days, abstracts are so marketing/advertising forward that it's hard to even understand the claim.
Re: NeurIPS 2025 Best Paper Awards
#18seems lots of chinese.
Re: NeurIPS 2025 Best Paper Awards
#19Interesting that 3 names I recognized as physicists from stat mech adjacent fields. They continue to punch above their expectations (as sampled by general dismissal of physicists in AI/ML on HN and reddit).
Do people not like physicists?
https://xkcd.com/793/ captures the stereotype well.