After Jacob Coxon's resignation and extinction warnings, a lot of people are asking 'how could AI possibly kill everyone?' and claiming AI safety researchers have no realistic answer.
This is false! Here are the 5 best scenarios I know of:
AI 2027: https://t.co/CaTNvafRI7 (I strongly recommend this one for being realistic, engaging, and if you dig into the appendices, highly detailed)
Paul Christiano's scenario (Former Head of Safety @ AISI, 2019): https://t.co/nkhTQJnjuS
Gwern Branwen's scenario (widely known independent AI researcher, 2022): https://t.co/WMmlDESf7g
Holden Karnofsky's high-level explanation (RSP Lead @ Anthropic, 2022): https://t.co/mAsLNdggQj
Joshua Clymer's scenario (ex-OpenAI, 2025): https://t.co/lJHJp6ErOx
(There's also the Sable story from https://t.co/JbZmSbog25, though you'll have to buy the book to read that one.)
Writing concrete, specific risk scenarios with enormous amounts of detail has been a major research project of many of the most prominent voices in the field! (With the current leaders in effort being https://t.co/YkSOj2urg4 and https://t.co/CaTNvafRI7)
@theo I used it to read images and get structured output. When I checked the cost per image processed I said "there must be a mistake this cannot be a fraction of a cent..." and YES IT IS!
@saulgcullen@levelsio Wow what a shit component. My Claude doesn't do this haha. I think the bar will heighten and people will know when an app is good or not.
@saulgcullen@levelsio I don't know how some people are saying that AI equalizes everyone when I spend a whole day with AI to develop a single feature. Because I discuss with AI, the quirks, the security holes, how to do it right and without that discussion and written plan the AI writes loads of shit.
@danushman@AISafetyMemes@tynervp For the majority of people this is true. But key people like scientists politicians and anyone who can genuinely influence AI future should not follow your advice lol
@BernieSanders This was a warning. And they plan to put similar AI's into robots soon.
What will happen with the test misfires with a robot? Will it kill someone because "the test went wrong"? If we don't put guardrails sure it will.
@ChronoDyneSys@BernieSanders You're the liar. The implied request was "IN THIS TEST COMPUTER solve this hacking test". And the AI said "I'm going to hack other computers to see if they hold the answer to my test". The decision of hacking outside the test network was 100% the AI's.
IT WENT ROGUE
@gilbert_jc@kimmonismus Just thought about this too. That paper seems it wasn't that far fetched. Maybe in the time-line but not what it predicts? If turns pretty extremist at the end but... this was also seen as science fiction a decade ago.
@robertocommit@ZssBecker Their apps are pretty bad. I'm frustrated with Claude app daily and I used Codex and was consuming soooo much CPU for simple stuff the UI was unresponsive.