organizational research article

Automated Weak-to-Strong Researcher

Anthropic · Published 2026-04

Read the original document ↗

Evidence from this source

Results and research linked to this document
TypeEvidence
ResultAutomated weak-to-strong research study · Nine Claude Opus 4.6 research agents
Source details

TL;DR, Section 3, Figure 1; best directed AAR chat-preference result

ResultAutomated weak-to-strong research study · Human-tuned weak-to-strong method
Source details

TL;DR, Section 1; strongest manually tuned chat-preference baseline

BenchmarkAutomated weak-to-strong research study
Source details

Article Sections 1–3, Figure 1 and associated caption

Paper / reportAutomated Weak-to-Strong Researcher
Source details

Primary publication source

Document tracking details
Source ID
src-automated-w2s-2026
Source type
Primary
Original URL
https://alignment.anthropic.com/2026/automated-w2s-researcher/
Retrieved URL
https://alignment.anthropic.com/2026/automated-w2s-researcher/
Last updated by publisher
Not reported
Access status
available
Retrieved
2026-09-28T17:53:28Z
Document fingerprint
c9c32a534f19f4055ae4248f0a2ede9000b0bbccda60fe12c48c7491fd37b3e9
Based on original bytes. Used to detect changes to the source.

Availability checks

  • 2026-09-28T17:53:28Z · success: Primary source inspected for this local candidate.