Open Source Models Write Better Than OpenAI/Anthropic, User Claims
JoshPurtell · x · 2026-08-20
User JoshPurtell argues that models from major labs are "atrocious" at writing. He observes that open-source models like Nemotron, which likely lack extensive writing-specific RLHF beyond DPO, write significantly better than proprietary models, questioning whether labs actually train for writing quality.
More from Models
- User complains Fable model repeatedly lies about work done — JoshuaJBouw · 2026-08-20
- Cartesia launches Sonic 3.6: 44-language TTS model with sub-90ms latency — emmanuelvivier · 2026-08-20
- Scaling laws for diffusion MoE released; LLaDA v2 rivals Qwen3 with 35% fewer tokens — jiqizhixin · 2026-08-20
- Uncensored ERP Review: Why Qwen3-235B-a22b Remains the Top Choice — redditaccountno6 · 2026-08-20
- Google's RT-2 packed 55B parameters, 1000x RT-1's 55M just 7 months later — binarybits · 2026-08-20
- Beating AI Detectors Is a Terrible Proxy for Training Good Writing Models — viemccoy · 2026-08-20