Development

A looped transformer trades parameters for iterations, leaking past compute caps

development2026-09-13

Claim

A recurrent looped transformer reuses its layers to think longer, trading parameters for iterations, exactly the kind of algorithmic gain that makes pacing by compute a leaky sieve.

Reporting date: 2026-09-13 (not necessarily the event date)

Sources

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

Welcome to September 13, 2026 (theinnermostloop.substack.com)

Knowledge graphopen in browser →

Reported in

Welcome to September 13, 2026

Evidence for

Supersedes

A recurrent-depth trick hides more of a model's thinking

Referenced by

Welcome to September 13, 2026

Cite or correct this record

Citation formatCitation text

A looped transformer trades parameters for iterations, leaking past compute caps
Innermost Loop KB — secondary record
Reporting date: 2026-09-13
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-09-13-pacing-by-compute-is-a-leaky-sieve
Claim: A recurrent looped transformer reuses its layers to think longer, trading parameters for iterations, exactly the kind of algorithmic gain that makes pacing by compute a leaky sieve.
Attributed sources:
- Welcome to September 13, 2026 — https://theinnermostloop.substack.com/p/welcome-to-september-13-2026
This citation does not imply independent verification of linked-source claims.

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