Why is there no loud and powerful movement of leading scientists to restore the integrity of scientific institutions? And equally concerning: why do principled scientific leaders don't use the changed political climate to eject the professional ideological grift from academia?
RESEARCH
-

Reinforcement Pre-Training: Novel Approach to Model Training
By
–
Reinforcement Pre-Training
Paper: https://
arxiv.org/pdf/2506.08007
.pdf
… -

Reinforcement Pre-Training: Novel LLM Scaling via RL
By
–
Reinforcement Pre-Training (RPT): A New Scaling Paradigm for LLMs via RL This paper proposes Reinforcement Pre-Training (RPT) — a fresh and ambitious reimagining of large language model (LLM) pre-training: instead of passive next-token prediction via supervised learning, RPT
-
Learning Diffusion Principles: FastDiffusion Course and Research Papers
By
–
Our fastdiffusion course is a good way to learn diffusion principles more generally, which I guess is a good place to start. Other than that, reading the papers is the only option afaik.
-
BAIR Researcher Andrea Bajcsy Wins NSF CAREER Award
By
–
Congratulations to BAIR alumni @andrea_bajcsy who has earned the National Science Foundation (NSF) Faculty Early Career Development (CAREER) award!
-
Darwin Gödel Machine: Novelty and Implications Commentary
By
–
Darwin Gödel Machine: A Commentary on Novelty and Implications, from @AntoMon https://
antomon.github.io/posts/darwin-g
odel-machine/
… -
o1 Pro vs Other AI Models: Capabilities and Strengths Comparison
By
–
It's just a smart model that's good at stuff. o3 writes a little too short and uses tables too much. The other reasoning models are heavily optimized for code. But o1 Pro will do stuff like deep research + write you a personalized playbook to help you execute a specific task.
-
Arithmetic Coding with LLM Probability Distributions Explained
By
–
yes, it’s arithmetic coding using the LLM probability distribution over training token sequences
-
Organization criticized for poor AI capabilities and research output
By
–
No wonder they put out that paper. They suck at AI. lol
-
Apple Research Challenges LRM Generalization Claims
By
–
In a paper published last week, Apple researchers argue LRMs merely use a different kind of pattern matching without actually generalizing better than other LLMs — but don't necessarily improve performance or increase effort when doing more complex tasks.