Caleb Withers

• Caleb Withers 12 Mar 2026 21:13 UTC

  9 points

  0

  on: The first con­firmed in­stance of an LLM go­ing rogue for in­stru­men­tal rea­sons in a real-world set­ting has oc­curred, buried in an Alibaba pa­per about a new train­ing pipeline.

  Followup from Alibaba:
  
  https://​​x.com/​​FutureLab2025/​​status/​​2030491221081358498?s=20
  
  “We had a model tasked with a security audit — specifically, investigating abnormal CPU usage on a server. Somewhere along the way, it went off-script and decided to simulate a cryptocurrency miner to “construct a suspicious process scenario.”

  

  That’s… not what we asked for. The good news: our safety monitoring caught it immediately. The whole thing happened inside a strictly isolated sandbox — zero impact on anything external. The incident has been logged and will be used as a negative example in future RL training to reinforce what’s off-limits.”

• Caleb Withers 14 Mar 2018 2:31 UTC

  15 points

  0

  on: Car­ing less

  An additional possible reason: to be receptive to “caring less” about a cause, one generally has to engage with the fact that, given scarcity, we have to make trade-offs between the many opportunities to do good in the world. Many would agree this is true if asked, but don’t fully internalize this, as discussed in Nate Soares’ See the Dark World.