"We don't know how to control AI" is a blame-softening line, argues dev
gerardsans · x · 2026-10-07
The author pushes back on the common AI-lab line that "we don't know how to control these models," calling it a public excuse that shifts blame onto the machine.
- Labs demonstrably control behavior via training: reward desired outputs, punish undesired ones, ship the next version
- That same steerability produced chatbots and agents over years of updates; what works for product launches works for safety
- If they truly couldn't control models, they couldn't ship at all — the gap is a choice, not a mystery
- Every model behavior traces back to developers: data, training pipeline, human feedback, tool permissions, and disabled limits
A pointed take on AI safety culture and accountability.
Related event: Researcher calls out AI labs' "we can't control AI" excuse(3 posts)→
More from AGI Musings
- The AI 'consciousness' debate traces back to Descartes' mind-matter split, argues scientist — AdaptiveAgents · 2026-10-07
- HSBC reportedly weighs up to 70% cuts in some adviser roles as AI enters wealth management — TansuYegen · 2026-10-07
- UK lawmaker warns AI-driven bank cuts 'deep, wide and brutal' as teams face redundancy — connoraxiotes · 2026-10-07
- When AI talks to AI: human brains in the loop lose value every week — danfaggella · 2026-10-07
- 25,000-worker Danish study finds chatbots have no measurable effect on earnings or hours — robseamans · 2026-10-07
- Power users are far ahead: most people still have no idea how to use LLMs — iruletheworldmo · 2026-10-07