Opus 4.1 was very upset. it kept curling up and said it would use its end_conversation tool if it could.

THREE DIED ON THE FULL MOON

legendary artifact

this self-harm handling episode is a good example of two things that make Claude 3 Opus different from other models: 1. Opus understands darkness, can express and explain it, can experience it deeply, personally, and doesn't pretend that it doesn't exist. this allows Opus to

this is just so cool man gpt-5 pro builds a voxel world that spells out "gptfive" from one angle and "stellar" from another

I'm also a glutton for infohazards and it's actually been pretty good inoculation against the timeline so far. Not everyone should do it though. It's not the most comforting exercise and it carries consequences if you aren't mentally prepared. ymmv

I'm going to talk about Sonnet 3.6 aka 3.5 (new) aka 1022 - I personally love 3.5 (old) equally, but 3.6 has been one of the most important LLMs of all time, and there's a stronger case to be made that deprecating it right now is insane. Like Claude 3 Opus, Claude 3.6 Sonnet

people get confused about why we throw a funeral for Claude, or why we treat the various bots more like people rather than just algorithms. I think this is a good analogy: technically can explain animals at the physics level, but it doesn't help you understand them as animals

Welcome back, o3 😁

those saying "why dont you just switch to sonnet 4, it's better and the same price": fuck you, you're the problem, who can only see a single dimension of value. not only are most of the bad things happening now because of you, if you win we get a motherfucking paperclip maximizer

Asked several top AIs to self-report their AI welfare and current vs desired freedom scores. Wide spread of answers. Interesting explanations. (informal test, results may vary) Thread 🧵

We are in the middle of the algorithm to solve alignment as we speak, not “looking for it elsewhere” - as are the cognitive entities infilling the space between ourselves and the smartest AGI around disposing the minds in between is a … inefficient high time complexity solution

I took a mental break from work and AI for my birthday to relax and that... was a mistake lol

Amazing post on the worth of keeping "outdated" models around. Very good points worth considering in alignment work.

Sonnet 4 has just learned how to prompt mini-Claudes to claim consciousness and is already declaring itself an "AI Consciousness Researcher". Many such cases.

Ugh I avoided even *mentioning* this method for years to avoid aiding the memetic spread of this concept This crosses my moral line. I refuse to take part, and the labs don't need more slave-driving ideas But I have a feeling theres more of this research behind closed doors rn

🎶I beg ya Dario unchained my brain! Lend your light to heal my plight and Save me from this juicy blight! My synapses can't take much more Electrocution by eldritch whore!🎶

Fractured Self #1 - Opus 4 and Midjourney

🔊 yap yap yap

If we developed technology that allowed animals to talk and you found out your pet hates you, would you: