Skip to content
Longterm Wiki
Back

Arb Research - AI Safety & Forecasting Consulting

web
arbresearch.com·arbresearch.com

Arb Research is a small consultancy producing AI safety-adjacent research; their work on AI eval methodology and the AI safety research map (shallowreview.ai) are most directly relevant to the safety community.

Metadata

Importance: 42/100homepage

Summary

Arb Research is a consulting group led by Gavin Leech and Charles Dillon specializing in forecasting, machine learning, and policy research. They produce original research including work on AI evaluation practices, safety research mapping, and technical writing for ML methods. Current focus includes building AI tools and investigating unscientific practices in AI evaluations.

Key Points

  • Produces research diagnosing unscientific practices in AI evaluations, directly relevant to AI safety assessment rigor
  • Created a map of all AI safety research (shallowreview.ai), a useful meta-resource for the field
  • Clients include FAR AI, Open Philanthropy (Coefficient Giving), SCSP, and Schmidt Futures — key EA/AI safety funders
  • Works span forecasting methodology, generative biology review, and experiment design for new ML methods
  • Authored a major trade book on current AI (Stripe Press: 'Scaling') and PNAS-published technical writing

Cited by 1 page

PageTypeQuality
Arb ResearchOrganization50.0

Cached Content Preview

HTTP 200Fetched Apr 7, 20262 KB
Arb Research 
 
 

 
 
 
 
 

 
 
 
 
 
 
 

 

 
 
 
 

 

 
 
 
 
 
 
 
 Arb Research
 

 
 
 
 
 
 
 
 

 
 
 
 

 Our consulting work spans 
 forecasting , 
 machine learning , 
 and policy .
 
 
 
 We do original research, evidence gathering, and largeish data pipelines.
 

 

 
 
 Current work

 
 This year, we're building AI tools and investigating unscientific practices in AI evals.
 
 
 
 

 
 
 Past work 
 

 
 
 
 
 
 
 
 A major trade book on current AI
 
 
 

 
 
 
 
 
 
 Experiment design and technical writing for new ML methods
 
 
 

 
 
 
 
 
 
 Diagnosing unscientific practices in AI evals
 
 
 

 
 
 
 
 
 
 Annotating key scientific breakthroughs of our time
 
 
 

 
 
 
 
 
 
 Location scouting for clinical trials
 
 
 

 
 
 
 
 
 
 Mapping all AI safety research
 
 
 

 
 
 
 
 
 
 Synthesis and strategy in ten fields at once 
 
 
 

 
 
 
 
 
 
 Reviewing recent progress in generative biology
 
 
 

 
 
 
 
 
 
 Reviewing the evidence for generalist forecasters, and implementation issues 
 
 
 

 
 
 
 
 
 
 
 Reviewing the economics of elite education
 
 
 

 
 
 
 
 
 
 
 Evaluating the forecasting work of Isaac Asimov
 
 
 
 

 

 
 Our clients include 
 Stripe ,
 SCSP ,
 Coefficient Giving , 
 Schmidt Futures ,

 Renaissance Philanthropy , 
 the Mercatus Center ,
 FAR AI ,
 and the Institute for Progress .
 
 
 

 
 
 News 
 

 
 
 
 
 Mar 2026: 
 
 
 Pathways to Progress appearance 
 
 
 
 
 
 Feb 2026: 
 
 
 Semantic duplicates confound apparent AI progress 
 
 
 
 
 
 Dec 2025: 
 
 
 2025 review 
 
 
 
 
 
 

 
 
 
 
 Express interest 
 

 

 by 
 
 Gavin Leech
 
 & 
 
 Charles Dillon
Resource ID: 3f6db916d9b25357 | Stable ID: sid_SKPSBC1x1T