Meta Hiring Pretraining Eval Researcher to Tackle Data Mixing
JesseDodge · x · 2026-08-08
Jesse Dodge from Meta announced a 1-year research role focusing on core LLM training evaluation challenges:
- Predictive metrics: Exploring what to evaluate during pretraining to maximize post-training performance.
- Cross-stage correlation: Identifying the most predictive evals across pretrain, midtrain, and post-train phases.
- Data mixing: Investigating optimal strategies for data composition.
More from Companies & People
- Zuck Pens Essay Arguing ASI Will Be Net Positive as Open vs. Pause Debate Heats Up — thursdai_pod · 2026-08-08
- At Stanford AASF, quote honors Jeff Dean: 'It's the ocean that matters' — edchi · 2026-08-08
- KPMG: Nearly Half of Executives Delay AI Agent Deployments as Costs Exceed Benefits — Polymarket · 2026-08-08
- Databricks Reveals Internal AI Cost-Cutting Playbook: Up to 90% Savings — pwendell · 2026-08-08
- 21-Year-Old Cook Self-Taught AI for 3 Years, Lands $84.5K Director Job — bralynn2222 · 2026-08-08
- UMD Releases Syllabus for 'Generative AI Agents' Course — furongh · 2026-08-08