Nathan Lambert
Who
AI researcher. Author of Interconnects (interconnects.ai) newsletter — focused on RL, RLHF, and AI progress. Former researcher at Hugging Face and AI2.
Key Views
- Skeptical of fast takeoff / recursive self-improvement narratives
- Coined “lossy self-improvement” (LSI) as contrast to recursive self-improvement (RSI)
- Believes complexity brake (Paul Allen’s term) will keep AI progress more linear than exponential
- Acknowledges current models are genuinely useful for knowledge work