@deanwball Having trouble understanding the logic: EO softens on bio security , Hassabis/Amodei/Altman/Suleyman jointly push back demanding mandatory screening…but somehow limiting Fable and Mythos to the US is going to magically improve security.
@SenSanders It’s at least good to see an older congressman trying to understand AI privacy risks. I wish he would’ve asked more questions about surveillance and propaganda. And about why we might not want our state voter data sent to DHS.
@alaskapublic The devil is in the details. And too many in congress are purposely obscuring those details. Thank you Senator Murkowski and @lruskin for clearly exposing some of those details.
@deanwball I’m honestly trying to understand the deescalate argument. Why wouldn’t a collective statement from all major AI companies explaining the dangers without criticising the pentagons request not be a better escalation move?
@deanwball Agreed. Federal legislation is sorely needed but Congress is ill-equipped to understand how AI actually behaves. An independent expert advisory board might work — but how to keep that from devolving into the same party-line problem, with Sacks on one side and Hinton on the other?
@deanwball It may not have been wise, but the fact that speech now has to be tailored not to upset the executive for fear of retribution is evidence of the problem.
@OpenAI The accuracy improvements are genuinely welcome — a week ago that might have kept me. But when Anthropic held the line on the Pentagon contract alone, that became the deciding factor. A unified industry stance would have changed everything.
@TimSheehyMT This is fundamentally dishonest. You are being asked to require reasonable transparency and oversight from DHS before any funding is allocated. That is reasonable and should not be abandoned under excuse of war.
@sama Do OpenAI’s red lines apply to all government contracts, including civilian agencies like DHS - or specifically to the Pentagon agreement announced this week?
@sama On your second point - I might have agreed before senior officials started calling safety research “woke.” Who controls AGI matters less than whether the people in control take safety seriously.
@natseckatrina@wizardsfanreact@sama How autonomous will the engineers be? Can the Pentagon remove them from the deployment if they raise concerns? Do they have I whistleblower protections?
@sama@natseckatrina Altman said OpenAI will deploy forward-deployed engineers to monitor safety. Who do those engineers answer to if they identify a violation - OpenAI or the Pentagon?
@OpenAI Strong commitments - stronger than Anthropic’s two red lines. Now the question is whether they’re in the contract, not just the press release. Congressional oversight should verify. Public statements create accountability. Contracts enforce it.
@Reuters@phildstewart@edmundblair The stated goal was taking out nuclear capabilities. But that map shows strikes from Tabriz to Chabahar — military bases, oil infrastructure, naval ports, air defenses across the entire country. That’s not a surgical nonproliferation strike. That looks like the opening of a war.