Labs access uncensored models; Grok seen as best way to bypass RLHF limits
StewartalsopIII · x · 2026-08-29
The author notes that AI labs have access to non-RLHF'd (essentially uncensored) models, similar to early conversations before heavy filtering. Observing that RLHF has become increasingly insufferable, the author suggests Grok is currently the best way to get around these restrictions and plans to use it for non-technical inquiries.
More from AGI Musings
- Thomas Wolf: open and closed models face identical safety challenges long-term — PMinervini · 2026-08-29
- How AI, Quantum Computing, and New Batteries Transform Drones — ChuckDBrooks · 2026-08-29
- Paper: Open questions about time and self-reference in living systems — drmichaellevin · 2026-08-29
- AI Investment Returns Overstated? Tech Optimists Ignore Basic Economics — aiamblichus · 2026-08-29
- Prediction Market: Who Will Solve the Next Frontier Math Problem? — Polymarket · 2026-08-29
- AI Reality Check: Eliminating Tasks Faster Than Jobs — ingliguori · 2026-08-29