"Accelerating RL Post-Training Rollouts via System-Integrated Speculative Decoding" Speculative decoding for RL rollouts! This paper speeds up post-training without changing the target policy’s sampling distribution. So a draft model proposes multiple tokens, and the policy
RESEARCH
-
ASI’s potential for permanent advantage via chip innovation
By
–
I agree those bottlenecks imply it won’t go foom overnight but an RSI-ing ASI may well invent better chip production process, better ways to use current chips, etc. in a way that adds up to permanent advantage.
-
Jack Clark Predicts 60% Chance of Self-Recursive AI Improvement Within 3 Years
By
–
Pay attention when @jackclarkSF says that there is a 60% chance of self-recursive improvement happening in less than 3 years time. That is the take off scenario – on a timeline that fits with all of Anthropic’s remarkably accurate predictions regarding AI capability development.
-
Nemotron 3 Super Tops Open Source EnterpriseOps-Gym Leaderboard
By
–
Benchmarks should reflect real-world performance. That’s why we’re excited to share that Nemotron 3 Super has topped the open source category on the EnterpriseOps-Gym leaderboard. This agentic gauntlet evaluates performance across 1,150 tasks in fully interactive environments
-
Clark’s RSI definition: frontier model trains successor, not human obsolescence
By
–
Note Clark’s definition of RSI here, from his newsletter, is “a frontier model is able to autonomously train a successor version of itself.” This is a weaker claim than what I assumed he meant, which was that human researchers would no longer be useful vs. AI ones.
-
Empathy as Hidden State Modeling in AI Agents
By
–
With empathy, do you mean the ability to model the hidden states of another agent?
-
Auditing CV Model Failure Modes Before Production Deployment
By
–
Accuracy is table stakes. Failure modes decide whether your CV model survives production.
— Satya Mallick (@LearnOpenCV) 4 mai 2026
Same benchmark scores. Opposite real-world performance.
Dr. Satya Mallick on what to audit before you ship 👇#ComputerVision #MachineLearning pic.twitter.com/USnlxQsp9kAccuracy is table stakes. Failure modes decide whether your CV model survives production.
Same benchmark scores. Opposite real-world performance.
Dr. Satya Mallick on what to audit before you ship #ComputerVision #MachineLearning -
HexRunner Achieves Stable 30 MPH Locomotion Through Engineering Design
By
–
Designing for Speed: How HexRunner Achieved Stable 30 MPH Locomotion
— Ronald van Loon (@Ronald_vanLoon) 4 mai 2026
by @lukas_m_ziegler
#EmergingTech #Engineering #ArtificialIntelligence #Innovation #Technology pic.twitter.com/1BRRejkfW4Designing for Speed: How HexRunner Achieved Stable 30 MPH Locomotion
by @lukas_m_ziegler #EmergingTech #Engineering #ArtificialIntelligence #Innovation #Technology -
Frontier model can autonomously train successor version
By
–
In his newsletter, he defines it as “a frontier model is able to autonomously train a successor version of itself,” which I admit is a weaker definition than what I assumed he meant (human AI researchers no longer useful vs. AI ones)
-
Jack Clark: 60% probability of RSI by end of 2028
By
–
Anthropic co-founder Jack Clark says 60% chance of RSI by end of 2028: