Tag archive
SDAR claims it beats GRPO and avoids the instability of naive GRPO+OPSD. Here's how you'd prove both - the three-way comparison, the stability instrumentation most reproductions skip, and an honest bill for running it. The series close.
What Changed Recent research from Hugging Face highlights a critical finding regarding the...
Forsy-AI/agent-apprenticeship treats completed agent work as portable 'experience compilations' you can install into later runs. Broad agent
Turing-winner Richard Sutton launched Oak Lab with a north-star goal of a trillion-parameter agent that learns and plans in real time on about 20 watts, betting on continual experiential learning over the static pre-train-then-freeze paradigm behind
Three new open research projects point the same way: instead of only learning what to do, agents are learning to simulate the environment itself, so they can practice in their own imagination.
A new paper pins the sudden collapse of multi-step tool-use training on runaway probabilities in a few control tokens, and shows that mixing in supervised examples stabilizes it.
Several research groups landed on the same idea at once - improve a model by learning from its own attempts instead of expensive human labels - and the field is debating whether it really removes the labeling burden or just hides it.
A new recipe applies the same reinforcement-learning approach that polished chatbots to an image generator, then merges separate skill models into one - improving how faithfully it follows prompts and edits.
Qwen-AgentWorld trains a model to simulate the environment an agent acts in, then uses that simulation as a cheap, controllable place to learn -- reporting gains beyond training in the real thing.
A top-ranked paper shows that in modern RL training, improving the model you optimize does not guarantee improving the model you actually deploy - because the two run on different engines that disagree on probabilities.
CompactionRL introduces a novel reinforcement learning strategy that integrates context summarization to help agentic LLMs overcome the limitations of finite context windows, significantly boosting performance on long-horizon tasks like coding. This breakthrough enables LLMs to effectively manage and learn from extended interaction histories.
A new proof shows that three widely used reinforcement-learning recipes for training reasoning models - GRPO, Dr. GRPO, and DAPO - are all just different operations on a single number, the spread of rewards within a group of sampled answers.