agientry.com/en/articles/...
agientry.com/en/articles/...
The agents got a GPT model to its target in 77% fewer training steps.
outerloop.science
The agents got a GPT model to its target in 77% fewer training steps.
outerloop.science
Code: github.com/csinva/weight-space-programming-autoresearch
「How we built our multi-agent research system」では、リードエージェントが複数のサブエージェントを並列に動かす構成が説明されています。
長時間動かす話や、ツールの作り方も公式記事があります。
ただ、ツール周りの助言は古くなっている部分がある、と注記されています。
Claude Code 101やKarpathyのautoresearchは、公式ページを確認できていません。
「How we built our multi-agent research system」では、リードエージェントが複数のサブエージェントを並列に動かす構成が説明されています。
長時間動かす話や、ツールの作り方も公式記事があります。
ただ、ツール周りの助言は古くなっている部分がある、と注記されています。
Claude Code 101やKarpathyのautoresearchは、公式ページを確認できていません。
Jeremy Qin et al.
#arXiv #cs.AI
Autonomous experiment loop extension for pi
Autonomous experiment loop extension for pi
makes me stupid giddy that my 3060ti is pegged while an ai mind trains various neural nets throwin stuff at the wall 2 see what works n doesn't :3
makes me stupid giddy that my 3060ti is pegged while an ai mind trains various neural nets throwin stuff at the wall 2 see what works n doesn't :3
the actually interesting part of research (causal investigation, literature grounding etc) is off limits for now
the actually interesting part of research (causal investigation, literature grounding etc) is off limits for now
#AndrejKarpathy #AI #learning
pythonlibraries.substack.com/p/from-100-l...
#AndrejKarpathy #AI #learning
pythonlibraries.substack.com/p/from-100-l...
+33% throughtput from baseline decode throughput
#171: Batch scattered writes to reduce memory contention (+16%)
#180: Keep workers fed with more blocks in flight (+10%)
#181: Eliminate speculation & replay (+3%)
+33% throughtput from baseline decode throughput
#171: Batch scattered writes to reduce memory contention (+16%)
#180: Keep workers fed with more blocks in flight (+10%)
#181: Eliminate speculation & replay (+3%)
“‘Scaling Discovery Requires Forgetting’. They ran autoresearch agents three ways and all plateaued for the same reason: ‘Each agent became anchored to its own past attempts, making fewer and smaller edits.’”
“‘Scaling Discovery Requires Forgetting’. They ran autoresearch agents three ways and all plateaued for the same reason: ‘Each agent became anchored to its own past attempts, making fewer and smaller edits.’”
Change "arxiv" to "autoarxiv" in any paper URL
OpenResearch creates a project for you, sets up an environment on your preferred compute, and runs a minimal reproduction. The agent tracks logs, visualizes runs, and produces artifacts, all entirely local
Change "arxiv" to "autoarxiv" in any paper URL
OpenResearch creates a project for you, sets up an environment on your preferred compute, and runs a minimal reproduction. The agent tracks logs, visualizes runs, and produces artifacts, all entirely local