Development

A small model is trained to acquire research taste rather than procedure

development2026-08-23

Claim

London's Inherent, founded by DeepMind alumni, said its research teammate Faraday, built on a 27B open base, beat Opus 4.8 and GPT-5.5 at reproducing published science, trained by reinforcement learning to acquire research taste rather than mere procedure.

Reporting date: 2026-08-23 (not necessarily the event date)

Sources

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

Welcome to August 23, 2026 (theinnermostloop.substack.com)

Knowledge graphopen in browser →

Reported in

Welcome to August 23, 2026

Actor

Evidence for

Supersedes

A 27B AI scientist beats far larger models at reproducing unseen figures

Referenced by

Cite or correct this record

Citation formatCitation text

A small model is trained to acquire research taste rather than procedure
Innermost Loop KB — secondary record
Reporting date: 2026-08-23
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-08-23-research-taste-trained-by-reinforcement-learning
Claim: London's Inherent, founded by DeepMind alumni, said its research teammate Faraday, built on a 27B open base, beat Opus 4.8 and GPT-5.5 at reproducing published science, trained by reinforcement learning to acquire research taste rather than mere procedure.
Attributed sources:
- Welcome to August 23, 2026 — https://theinnermostloop.substack.com/p/welcome-to-august-23-2026
This citation does not imply independent verification of linked-source claims.

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