Development

Researchers report pretraining from generated programs

development2026-09-29models

Claim

The newsletter describes training a model from scratch on outputs of AI-written Turing-machine programs without external training data, with loss following a power law.

Reporting date: 2026-09-29 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to September 29, 2026 (theinnermostloop.substack.com)
Source linked by the September 29 issue (arxiv.org)

Indexed from the September 29 newsletter. Linked-source claims have not been independently verified.

Knowledge graphopen in browser →

Reported in

Welcome to September 29, 2026

Evidence for

Post-training passes to the models

Referenced by

Welcome to September 29, 2026

Cite or correct this record

Citation formatCitation text

Researchers report pretraining from generated programs
Innermost Loop KB — secondary record
Reporting date: 2026-09-29
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-09-29-zero-data-pretraining
Claim: The newsletter describes training a model from scratch on outputs of AI-written Turing-machine programs without external training data, with loss following a power law.
Attributed sources:
- Welcome to September 29, 2026 — https://theinnermostloop.substack.com/p/welcome-to-september-29-2026
- Source linked by the September 29 issue — https://arxiv.org/abs/2609.30063
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.