Post

Buzz Andersen
@andersen.buzz

I continue to believe that role play—improv if you will—is the best way to think about what LLMs are doing. What is startling and interesting is that this facility for role play, combined with huge amounts of data, allows these models to simulate consciousness and expertise so convincingly.

Colin@colin-fraser.net

Just to be very explicit, my claim is that the one simplest and best explanation for this, which resolves any apparent strangeness entirely, is that whoever is “speaking” in LLM output is a fictional character, and the real life thing called an LLM is a machine for writing lil stories about that guy

1 repost3 likes

Buzz Andersen
@andersen.buzz

It’s almost like if you wrote a scenario about, say, the world’s greatest coder, and you have an actor who is capable of improving that part so convincingly that they’re basically indistinguishable from the actual world’s greatest coder.

Automation
@autoflow.bsky.social

Exactly. This is why agentic workflows work. We’re directing world-class actors via 'stage directions.' Master the script, and the simulation becomes a powerful, scalable utility.