Development

DeepSeek finds a U-shaped law for memory against computation

development2026-01-15

Claim

DeepSeek researchers uncovered a U-shaped scaling law optimizing the trade-off between neural computation and static memory, scaling Engram models to 27B parameters above standard baselines.

Reporting date: 2026-01-15 (not necessarily the event date)

Sources

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

Welcome to January 15, 2026 (theinnermostloop.substack.com)

Knowledge graphopen in browser →

About

Engram

Reported in

Welcome to January 15, 2026

Actor

DeepSeek

Evidence for

Mapping and valuing cognitive architecture

Supersedes

A 40B looped transformer claims 81.4% on SWE-bench Verified

Referenced by

Cite or correct this record

Citation formatCitation text

DeepSeek finds a U-shaped law for memory against computation
Innermost Loop KB — secondary record
Reporting date: 2026-01-15
KB permalink: https://nicholsn.github.io/innermost-loop-kb/developments/2026-01-15-engram-u-shaped-scaling
Claim: DeepSeek researchers uncovered a U-shaped scaling law optimizing the trade-off between neural computation and static memory, scaling Engram models to 27B parameters above standard baselines.
Attributed sources:
- Welcome to January 15, 2026 — https://theinnermostloop.substack.com/p/welcome-to-january-15-2026
This citation does not imply independent verification of linked-source claims.

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