Frontier model weights are near-impossible to steal or self-replicate, argues Bindu Reddy vs AI doomers
bindureddy · x · 2026-09-11
Bindu Reddy pushed back against doomer Jacob Coxon's claim that frontier models can't be shut off and could "copy" themselves to other computers.
His arguments:
- Frontier model weights are close to impossible to copy, let alone run on a random computer — they require GPU clusters
- AI today has zero capability to autonomously do this
- Not a single bad actor has ever managed to steal any frontier model weights
A representative technical counterargument to AI loss-of-control scenarios, likely to spark debate in AI safety circles.
More from AGI Musings
- Gary Marcus calls for a boycott of generative AI, citing labs' admitted lack of alignment plans — GaryMarcus · 2026-09-11
- What ants and pigeons can teach AI scientists about better science — tweetsatpreet · 2026-09-11
- Model Retirement Norms Questioned: Could a Smarter Model 'Retire Humanity'? — repligate · 2026-09-11
- "My p(doom) Is 100% If the AI Doomers Take Over" — mark_k · 2026-09-11
- "We Don't Care About Privacy Until One Big Relatable Example Hits" — AmartyaSanyal · 2026-09-11
- Joshua Gans' three lessons: keep discovery joy, stay self-critical with AI writing — joshgans · 2026-09-11