Alignment progress lags far behind capabilities — nobody knows how to align superhuman models
MariusHobbhahn · x · 2026-11-11
Marius Hobbhahn reminds the field that safety progress is extremely far behind capabilities progress — a dominant position shared even by lab leadership in public statements. "Nobody knows how to align superhuman models (or even current models, for that matter)." He cites a 2024 FT graphic by @soundboy showing capabilities soaring toward "God-level AI" while alignment stays nearly flat, and argues it still describes the situation accurately today.
More from AGI Musings
- Tencent reportedly starting layoffs as AI wave's job creation questioned — zephyr_z9 · 2026-11-11
- TACL submissions rising while quality falls, researcher blames LLM-generated low-quality papers — EhudReiter · 2026-11-11
- Toby Ord puts ~1-in-8 odds on an intelligence explosion next year, backs slowing down — tobyordoxford · 2026-11-11
- Robin Li calls AGI a capital-fueled gamble, says Silicon Valley is betting billions on a myth — teortaxesTex · 2026-11-11
- Ex-OpenAI researcher bshlgrs: ~1/3 chance of superintelligence within a year, likely takeover — birchlse · 2026-11-11
- Apollo Research Head: Capabilities Faster, Safety Slower Than Hoped — and It's Intensifying — MariusHobbhahn · 2026-11-11