StackMap
Subscribe

autoresearch vs sia

Karpathy's autoresearch loop as an installable skill for Claude Code, OpenCode and Codex: constraint + mechanical metric + autonomous modify→verify→keep/discard iteration. — versus — Self-improving loop from the SIA paper: Meta, Target and Feedback agents evolve a task agent's harness AND weights against a benchmark — #1 on MLE-Bench Hard, 14x kernel speedups.

The curated verdict

Same lineage — the autoresearch loop: metric, modify, verify, keep or discard. The skill hill-climbs your code with a fixed agent; SIA makes the agent itself the thing that improves, up to and including its weights.

autoresearchsia
Stars5.4k2.1k
Forks399244
LanguageShellPython
LicenseMITMIT
Last activity1 months ago21 days ago
Topicsskills, codingagents, training
Curated connections42

autoresearch — the curator's take

The compounding-gains loop, packaged: pick a metric a machine can check, let the agent mutate-verify-keep against it for hours, and small wins stack — the Karpathy recipe without writing the harness yourself. Works across three agent CLIs. The discipline it demands is the catch: without a truly mechanical metric the loop optimizes noise, and unattended iteration burns real tokens — set budgets before you set it loose.

sia — the curator's take

The most ambitious of the self-improvement loops on the map: where others mutate prompts or code, SIA's Feedback agent rewrites the task agent's harness and updates its weights, with paper-grade receipts (70.1% on LawBench vs 45% prior SOTA, 14x on an AlphaFold-3 Triton kernel, #1 on MLE-Bench Hard). Reach for it when your problem IS a benchmark: a scoreable task the loop can grind against. NOT for tasks without a mechanical metric, and not a polished product — it's the official research implementation, so expect to build the task harness yourself and budget real GPU time for the weight-update path.