bearseascape

8 ms·

1.
A somewhat optimistic view of AI in mathematics(proofsandprompts.com)
1 points by bearseascape 8d ago|0 comments
2.
by bearseascape 9d ago
They did do this, but it seems that Navier-Stokes is the only one that was successful (unless, for some strange reason, they have proofs for the other problems but haven’t released them yet).
3.
To Serve Man: AI, Math, and Navier–Stokes(ml5885.github.io)
4 points by bearseascape 9d ago|1 comments
4.
by bearseascape 9d ago
I wrote this to try and contextualize the recent developments around the Navier-Stokes equations with the broader debate ongoing in mathematics about the role of such AI-generated proofs. I'm not a mathematician, and there are probably
5.
Model Spec Midtraining: Improving How Alignment Training Generalizes(alignment.anthropic.com)
2 points by bearseascape 5mo ago|0 comments
6.
Following the Text Gradient at Scale (2025)(ai.stanford.edu)
9 points by bearseascape 5mo ago|1 comments
7.
Transformers Are Inherently Succinct (2025)(arxiv.org)
62 points by bearseascape 5mo ago|9 comments
8.
Slople – Can you tell real ML papers from AI-generated ones?(ml5885.github.io)
3 points by bearseascape 6mo ago|1 comments
9.
by bearseascape 6mo ago
I was inspired by https://slop.zackg.me/ , which does the same thing for PL papers but with entire fully generated AI papers (i.e. LaTeX-rendered PDFs and all) - check it out! I thought it was fun to play, and wanted a versi
10.
Benchmarking Culture(argmin.net)
1 points by bearseascape 6mo ago|0 comments
11.
by bearseascape 8mo ago
> What counts as research? You might be aware of this, but most big tech companies (i.e. the ones with massive user counts) don't just let you roll out UI changes to everyone, because they know that this has a downstream impact on u
12.
Why one small American town won't stop stoning its residents to death(archiveofourown.org)
2 points by bearseascape 8mo ago|1 comments
13.
by bearseascape 8mo ago
A parody of a fictional New Yorker article, in which journalist Isaac Chotiner grills the administrator from Shirley Jackson's "The Lottery" about the civic benefits of ritual stoning.
14.
The most complex model we understand [video](youtube.com)
2 points by bearseascape 9mo ago|0 comments
15.
Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs(arxiv.org)
1 points by bearseascape 9mo ago|0 comments
16.
by bearseascape 1y ago
The madlib like sentences approach is actually how masked token prediction works! It was one of the pretraining tasks for BERT, but nowadays I think all (?) LLMs are trained with next token prediction instead.
17.
MooseAgent: A LLM Based Multi-Agent Framework for Automating Moose Simulation(arxiv.org)
13 points by bearseascape 1y ago|0 comments
18.
Automated Researchers Can Subtly Sandbag(alignment.anthropic.com)
2 points by bearseascape 1y ago|0 comments
19.
Auditing Language Models for Hidden Objectives(anthropic.com)
1 points by bearseascape 1y ago|0 comments
20.
Policy for LLM Writing on LessWrong(lesswrong.com)
2 points by bearseascape 1y ago|0 comments
21.
Towards Understanding Distilled Reasoning Models: A Representational Approach(arxiv.org)
3 points by bearseascape 2y ago|0 comments
22.
Transformers Learn to Implement Multistep Gradient Descent with Chain of Thought(arxiv.org)
1 points by bearseascape 2y ago|0 comments
23.
(Mis)Fitting: A Survey of Scaling Laws(arxiv.org)
2 points by bearseascape 2y ago|0 comments
24.
Resurrecting saturated LLM benchmarks with adversarial encoding(arxiv.org)
1 points by bearseascape 2y ago|0 comments
25.
Deep Double Descent: Where Bigger Models and More Data Hurt(openai.com)
2 points by bearseascape 2y ago|0 comments
26.
Value-Based Deep RL Scales Predictably(arxiv.org)
68 points by bearseascape 2y ago|3 comments