Karpathy’s autoresearch repository sets AI agents loose on single-GPU nanochat training runs, proposing, running and scoring modifications without a human in the loop (announcement); the next issue would record 650 experiments in two days. Alongside it, Codex 5.4 (Codex) took a DOS game from raw binary to a Rust rebuild in hours, unpacking assets, disassembling the executable and rewriting the renderer (thread). The research loop formalizes the December moment when Karpathy handed his nanochat optimization to Claude, and it seeds the lineage the corpus follows through Bilevel Autoresearch and the 72-hour, 50-experiment run in April.
Development
A model reverse-engineers a DOS game from raw binary into Rust
The corpus's autoresearch lineage begins here, with a researcher handing the experiment loop itself to agents, paired with a reverse-engineering feat showing coding agents finishing long, multi-stage builds end to end.
Claim
Karpathy's autoresearch project autonomously conducts training research on language models, while Codex 5.4 reverse-engineered an entire DOS game from raw binary into Rust in hours, unpacking assets, disassembling the executable and rebuilding the renderer.
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to March 8, 2026 (theinnermostloop.substack.com)
- Karpathy announces autoresearch (x.com)
- Codex 5.4 reverse-engineers a DOS game from raw binary into Rust (x.com)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.