@at_fome You cant say this without acknowledging the progress. Last December, the best models were able to solve some novel but very niche open math problems. Now, merely a year later, a millennium prize problem has fallen to an AI. Look ahead, what do you think Claude will do next year?
@elshayib_ You can't have an AI model you plan to release and use and just keep it in a box like this forever. The usefulness increases significantly even with light internet access
@Devgymsetvfx Is it hard to understand that slowing down AI for a few years to significantly reduce X risk is obviously a good idea for people with P(doom) > 10%