AI writing latency stagnant; RL focus may degrade quality
peterwildeford · x · 2026-08-25
Author notes the lack of good evals leaves us flying blind on important progress. The 'time horizon' for AI writing relative to human interaction (needs to be <5min) has barely increased, arguably worsened by a strong focus on RL.
More from Models
- $40 experiment: Opus 5 hits ~0.94 F1 on ExtractBench, Qwen3.8 matches at 1/3 the price — Ok-Challenge-7810 · 2026-08-25
- Claude vs. Grok: Restrictive safety vs. clever workarounds — whatsallthiss · 2026-08-25
- 165 GPU Hours Testing 12 Abliterated Gemma 4 12B Variants: The Most Jailbroken One Destabilizes Reasoning — nathandreamfast · 2026-08-25
- OpenAI's new inference engine boosts throughput up to 4.1x on select models — rohanpaul_ai · 2026-08-25
- OpenAI's Codex lead: next-gen models need more than your laptop, Codex will look primitive — koltregaskes · 2026-08-25
- AI4Bharat's 0.8B IndicOCR Takes On Models 4-5x Its Size, Open Weights — sumanthd17 · 2026-08-25