Loading activity…
@youjiacheng
AI
Antonia Bronars and colleagues report vision-free assembly; their abstract provides real-world results only for peg insertion.
AI
One commenter says it's hard to separate distillation's contribution from gains made through internal reinforcement learning.
AI
A user says those checkpoints span both the ladder and the big run.
AI
The paper’s author says coding agents can run for days on one problem but sometimes stop improving early.
AI
The authors say they use Fisher information to connect theory with empirical findings across RL, SFT and on-policy distillation.
AI
A user links a PDF they describe as the roadmap and attributes its release to copyright litigation.
AI
A team member says one policy learned using a binary success-or-failure reward and a simple control cost.
AI
A user argues that strong conditioning needs to accompany large data.
AI
A user ties the potential increase to data repetition; overfitting means fitting training examples too closely.
AI
The author says their approach scales with compute without hand engineering and learns fast, reactive policies without imitation data.
AI
A user raises the question in a quote post that links to another account's post.
AI
A user quote-posted another post and praised an unspecified action without giving further details.
AI
A user asks whether a linked claim is mathematically true if real-number arithmetic is assumed.
AI
A developer reports 20–28 millinats lower final validation loss than Muon across tested model sizes.
AI
Its creators say the method has an LLM recover a future paper’s idea, charging each hint by how surprising it is.
AI
A user quotes the line from another post and says, “wow that makes sense.”
AI
Its team claims activation perturbations can approach backprop, but an outside researcher questions the evidence.
AI
The poster said the boost covers subscriptions and Sign in With ChatGPT partners, with no changes needed from users.
AI
A user revised an earlier thought that a model might need a large mean key, saying it was “still optimization.”
AI
A commenter says rewriting training data to optimize for surprise where it matters gives backpropagation a cleaner signal.
