ComfyUI-Copilot: An Intelligent Assistant for Automated Workflow Development ComfyUI-Copilot is an LLM-powered plugin designed to simplify and accelerate workflow creation in ComfyUI, an open-source AI art platform, by providing intelligent node/model recommendations and
LLMS
-

AReaL: Asynchronous Reinforcement Learning System for Language Reasoning
By
–
AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning AReaL is an asynchronous reinforcement learning system that efficiently trains large language models for reasoning tasks by maximizing GPU usage and decoupling generation from training.
-

ProRL: Long-Horizon Reinforcement Learning Unlocks New LLM Reasoning
By
–
ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models This paper introduces ProRL, a method that uses long-horizon reinforcement learning to unlock new reasoning strategies in LLMs—strategies that base models cannot access, even with
-

Negative Reinforcement Improves LLM Reasoning Without Explicit Rewards
By
–
The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning This paper shows that punishing incorrect answers—without explicitly rewarding correct ones—can be surprisingly effective for improving reasoning in large language models trained via reinforcement learning
-

Chain-of-Thought Prompting: Constraint Mimicry Over True Reasoning
By
–
CoT is Not True Reasoning, It Is Just a Tight Constraint to Imitate: A Theory Perspective This paper challenges the idea that Chain-of-Thought (CoT) prompting enables true reasoning in LLMs, arguing instead that CoT acts as a structural constraint that guides models to imitate
-

Gradient Surge in LLM Training: Weight Decay and Normalization
By
–
Why Gradients Rapidly Increase Near the End of Training This note investigates a sudden rise in gradient norms during the late stages of LLM training and identifies a surprising cause: the interplay between weight decay, normalization layers, and scheduled learning rate decay.
-

High-Entropy Tokens Drive LLM Reasoning via Reinforcement Learning
By
–
Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning This paper investigates how a small subset of high-entropy tokens—termed "forking tokens"—drives the performance of reinforcement learning with verifiable rewards (RLVR)
-
Autonomous AI: Evolution Without Human Intervention
By
–
Et si une IA pouvait s’auto-modifier, se corriger, évoluer… sans aucune intervention humaine ? C’est l’idée radicale explorée par Sakana AI : une machine qui code sa propre évolution.
— VISION IA (@vision_ia) 9 juin 2025
Forget fine-tuning — bienvenue dans l’ère des IA autonomes. pic.twitter.com/iFYt74mDygWhat if an AI could self-modify, self-correct, and evolve—without any human intervention? That’s the radical idea explored by Sakana AI: a machine that codes its own evolution. Forget fine-tuning—welcome to the era of autonomous AI.
-

Gemini 2.5 Pro ‘Deep Think’ Feature Available via API
By
–
Oh? "Deep Think" on Gemini 2.5 Pro is already available via API to some early access users, it seems. The post also mentions that it will arrive at AI Studio soon! Deep LFG h/t @atoz51515640
-
Input Token Proportionality in Language Model Economics
By
–
no…. they are both proportional to the number of input tokens…
