Industry voice: RL environments are largely generated at scale by closed models, not human labs
xeophon · x · 2026-09-25
In a thread about the AI data-labeling industry, @xeophon adds a corrective point: much of the data industry is actually generating environments at scale with closed models and using them heavily for training, rather than relying on full human-lab annotation. He notes this holds at least in the domains currently in focus (coding, cyber), making the industry picture messier.
More from Models
- opencode makes $60/month of DeepSeek v4.1 Flash usage permanent — intellectronica · 2026-09-25
- Low-Quality Photo LoRA Test: Flux1-dev Still Beats Krea 2, Z-Image and Flux2 — wreck_of_u · 2026-09-25
- Mk1.5 real-time multimodal reasoning model ships with 2-5x lower latency than Mk1 — AkshatS07 · 2026-09-25
- Dev swaps Gemini 3.1 for Jev as LLM judge: 200x cheaper, 50x faster, same accuracy — indragie · 2026-09-25
- Claude Opus 5.5 Turns a Home Video Into a Furnished 3D Zillow Listing — socialwithaayan · 2026-09-25
- 48 Hours of Claude Opus 5.5: Games, Films and a LEGO Set Already Shipped — socialwithaayan · 2026-09-25