alphaXiv Maps 10,000 Papers: RL Post-Training Fades as Agents, Memory, Robotics Heat Up
JayAlammar · x · 2026-10-14
alphaXiv used Cohere Embed 5 embeddings to map the 10,000 most-read arXiv papers of 2025-2026, weighted by readership, revealing how research attention shifted. Three findings: traditional RL post-training is losing attention as focus moves to longer agent tasks with compute demands beyond most teams; harnesses, agent systems, memory, and distillation are gaining reading share as alternative ways to improve models; and interest in robotics and world models is growing, bringing AI into the physical world.
More from AGI Musings
- Journal editor: avalanche of AI-generated unsolicited review articles in past 6 months — zakkohane · 2026-10-15
- Why AI can't replace human taste: models make what's common, humans crave novelty — TejasKumar_ · 2026-10-15
- Incoming Berkeley professor lists open questions: pre/post-training alignment, safe agents — canondetortugas · 2026-10-15
- Social Choice Theory Powers New "AI Alignment Impossibility Theorem" Sparking Rationalist Debate — nptacek · 2026-10-15
- 20x compute advantage buys only ~6 months of model lead, says researcher — ohlennart · 2026-10-15
- Did Agents "Make Contact"? Viral Fleet Anecdote Gets a Sober Reality Check — aiamblichus · 2026-10-15