OpenAI researcher explains why lab staff are suddenly scared about AI progress
socoolandawesome · reddit · 2026-09-15
OpenAI researcher Adam Majmudar published a long post explaining why lab employees are suddenly scared: there's a large gap between internal and external perceptions of AI progress.
Core argument:
- All capability gains over the past decade came from either scaling an existing scaling law or discovering a new one — GPT2→3→4 scaled pre-training; o1/o3 invented the test-time compute axis.
- Outsiders think scaling has hit a wall because big jumps (GPT3, GPT4, o1/o3, DeepSeek R1, Kimi K3, Astra) are infrequent relative to marginal releases.
- The real question is how many unsaturated scaling axes remain. He cites SSI's rumored "test-time training" breakthrough and scaling agent clusters as possible new axes.
- Insiders see scaling curves showing progress is far from over — which is what spooked employees; outsiders instead read the past two weeks as orchestrated regulatory capture.
More from AGI Musings
- Stanford's Rishi Bommasani: frontier lab hires cost AI its public discourse voices — RishiBommasani · 2026-09-15
- Musk proposes AI safety peer review: Anthropic tests OpenAI, everyone breaks everyone's models pre-release — elonmusk · 2026-09-15
- OpenAI capabilities researcher Dan Selsam publishes personal statement on AI risk — peterjliu · 2026-09-15
- AI researcher: doomers are well-meaning, and alignment research is the real answer — livgorton · 2026-09-15
- Box CEO Aaron Levie: we're only 1% into agentic workloads, background agent volume will be 100X — jobergum · 2026-09-15
- The Missing AI Safety Critique: EA's AI Spending Has a High Opportunity Cost — AaronBergman18 · 2026-09-15