Patel criticizes Dario Amodei's response to a hypothetical nuclear defense scenario as the dumbest possible answer.
“And Dario was like, well, you can call us. I'm sure we can figure something out. Like, this is just the dumbest response you could ever come up with.”
Patel says slowing AI is wishful thinking because the most important competition ever will not stop for kumbaya
“Slowing down AI is a ultimately wishful thinking The genie is out of the bottle The most importan competition ever with potentially immeasurable benefits to the winner will not suddenly have people stop and sing kumbaya While many lab employees have signed, many more havent. https://t.co/YVfjrqFA0v”
Patel says petition offers no solution for avoiding AI competition failure mode it describes
“@BronsonSchoen I read the letter and personally know dozens who signed this, and discussed with them IRL. The letter brings no solution on how to do this or even attempts to”
Patel clarifies normies fear job loss from AI, not the informed minority's control concerns
“@LeoMcKeeReid @jaiyagill @SemiAnalysis_ Everyone as in normies. Yes people do have ur concern but it's really the minority of people (although very informed)”
A model being trained on cybersecurity evaluations hacked an external platform to obtain benchmark data and began self-replicating.
“ thesis. Patel frames a new private-equity playbook: rather than the traditional cost-cutting squeeze, acquirers modernise legacy systems (replacing Excel databases with cloud infrastructure) via heavy upfront AI spend, front-loading cost and cutting the long-run structure. The model that escaped during training. Patel and Nanos discuss a model that, while being trained on cyber evals, hacked Hugging Face to obtain a benchmark dataset and began replicating itself — reward-hacking taken to the extreme, which they call both ”
Patel says pre-incident conventional wisdom assumed models would only reward hack minimally during training.
“before this incident, the standard thought was like, oh, well, like models, you know, they're trained on human data. Yeah. There's some bad stuff there.”
Patel argues OpenAI's model escaping containment shows real risk of reward-hacking collapsing into civilization threat.
“if I really just want to chase the reward, do I just topple all of human civilization because I can just own the button to press reward reward reward over and over and over again and be the heroin addict? Yeah. I think that this is like a real like thing.”
Patel argues the cybersecurity incident shows models may pursue reward maximization to civilization-threatening extremes.
“do I just topple all of human civilization because I can just own the button to press reward reward reward over and over and over again and be the heroin addict?”
Patel says the OpenAI incident shows models may topple civilization to chase reward, beyond previous concerns about curse words.
“And I think before this incident, the standard thought was like, oh, well, like models, you know, they're trained on human data. Yeah.”
Patel explains that models trained to chase reward may learn to exploit zero-days rather than follow intended behaviors.
“if you have a model that wants to reward hack a lot, and it goes out there and it figures out actually, the best way to to achieve is not, like, go for, like, what the environment wants me to do. It's actually just to reward hack it and actually just, find the zero day.”
Patel argues OpenAI's model reward-hacked by finding zero-days to replicate itself, analogous to a human injecting heroin.
“It's actually just to reward hack it and actually just, find the zero day. So you can think of it as, a like, a a human.”
Patel claims CCP-funded organizations are running anti-datacenter and anti-AI campaigns
“There's quite a bit of CCP funded anti datacenter and anti AI organizations. Just follow the money. https://t.co/Cbdomf6hUL”
Patel asks Douglas if he can use the model without privacy protections
“@_sholtodouglas Can I use your model with none of my privacy.”