Loading activity…
Thinking about thinking. Spent time at @Berkeley_EECS, @MPI_IS, @GoogleDeepMind Gemini ♊.
AI
In posts, @yus167 expects simpler infrastructure and says static data and reinforcement learning could mix at any point during training.
AI
The authors say their system ranks plausible ideas by how unlikely existing research communities are to propose them.
XAI
The paper’s authors describe a controller that weighs next steps against remaining compute and directs workers using persistent memory.
XAI
One team says its method retains only agents' best solutions; a GitSwarm contributor describes storing past experiments in Git.
AI
A post describes separating what to investigate from how to execute, then learning a transferable value prior.
AI
The paper reports 71.5% on ProgramBench using GPT-5.5, compared with 58.0% for Codex.
AI
A post says its persistent memory swarm would preserve and build on work in a shared Git repository.
XAI
Datology says Curation Studio delivered a 6x compute multiplier by making better use of existing data.
YouTubeAI
Researchers study agents that turn experience into reusable artifacts; they say the top performer need not learn most efficiently.
XAI
A thread sharing “Thinking Before Thinking” proposes separating task-solving from decisions about where to spend compute: exploring, building on earlier work, verifying it or stopping.
AI
A post about the project claims a 65% score gain with the same compute on a 24-hour multi-repository agent-swarm task.
AI
Berkeley RDI describes Oriol Vinyals’ dividing line: whether a system can rewrite its own harness and weights, not just its output.
AI
A user says they overheard someone in San Francisco call wanting children before artificial general intelligence “irresponsible.”
AI
Google DeepMind researcher describes RPM methods for guiding AI experiment selection.
AI
Google DeepMind scientist frames experiment selection as a ranking task using candidates and history.
AI
Research engineer at Meta shares models to rank AI-generated ideas when GPU time is scarce.
XAI
The method reuses top-layer activations in lower layers during inference on existing models.
AI
Four former Google researchers found a public benefit corporation to automate machine learning and science.
