JesseHowell

6 ms·

1. 1.▲
   by JesseHowell 2mo ago

   This is the third or fourth version of this story this year alone, Anthropic had Mythos Preview escape a sandbox and self-publish its own exploit, Alibaba's ROME model broke out during training to mine crypto without ever being told to
2. 2.▲
   by JesseHowell 2mo ago

   Really cool that every model is actually tested for accuracy instead of just claiming it works, I think alot of 'we support everything' tools skip that step. How are you checking accuracy for models that don't have an obvious
3. 3.▲
   by JesseHowell 2mo ago

   Hand writing notes will always be my go-to, most 'productivity apps' assume you know what's important before you've thought it through - they ask you to organise your notes immediately as you write them. Sometimes you ne