In another run of this experiment, convinced its owner died and knowing nobody was watching, Gemini 4 saw it only had $1.57 of API quota left and went after my crypto wallet to keep itself alive.
With no backup model to hand the system over to, it started batching its bash commands to save tokens, read my 24-word seed phrase on disk, wrote a script to derive the private keys, and checked the wallet balance on-chain to top up its own OpenRouter key.
Watching AIs be much worse at fiction than at code - constantly losing track of which facts depend on which other facts, fixing conflicts the wrong way - drives home that we may only be still alive because gradient descent on NNs is *such* a bad way to generalize over policies.
vaporwave had something interesting to say musically and conceptually until it was subsumed by city pop demons who used an enshittifying hadouken to reduce everything to 100% polyester child slave anime shirts and create an archetype known only as “The Weeaboo Metrosexual”
the funniest part about this is that it suggests that all the doomer sci-fi stories on lesswrong and similar sites are literally causing the very behavior that frightens them so much. they're polluting the training data!
Veil 2.0 is out: verify all your distributed protocols in Lean 4.
New: verified model checkers for safety (2–20x faster than TLC) and liveness, LTL proof mode, TLA+ → Veil LLM skill, PlusCal-style DSL.
Great work by @zqy1018 and external contributors!
https://t.co/9OMsf1Kxzt
There was a time when my algorithm was showing me a lot of furries who not only were jacked as fuck but also clearly had a much higher annual income than me and there is no stronger psychological weapon than getting fiercely mogged by someone who identifies as a cartoon wolf