poolside researchers on turning tens of thousands of data-mix experiments into frontier models
arena · x · 2026-08-21
LMArena's latest Conversations episode: Peter Gostlev talks with two poolside researchers, Connor Adams and Ravi, about learning fast through iteration—turning tens of thousands of experiments across data mixes into models that move the frontier. Topics include rapid iteration, massive pre-training runs, and RL decisions. Full video on Arena's YouTube channel.
Related event: poolside Researchers on Scaling Data Mixes for Frontier Models(2 posts)→
More from Models
- Google: Gemma 4 31B Matches Sonnet 5 Quality at 40x Lower Cost — 0xsachi · 2026-08-21
- Direct-OPD Transfers Weak Model RL Gains to Strong Models — tokenbender · 2026-08-21
- Prompt caching cuts consistency-based hallucination detection from 6x to ~1.5x cost, measured across 6 models — Own-Establishment293 · 2026-08-21
- New Chinese model OX Alpha spotted, Zhipu and Xiaomi suspected — vista8 · 2026-08-21
- Discussion: if frontier labs keep models internal, do open weights still have a path forward? — PykeTheTitan · 2026-08-21
- Ox Alpha: Free Chinese model with 1M context but high latency — bindureddy · 2026-08-21