nk

Good luck to you. I try to figure out how to prime it also, it's a bit of work and not easy to gauge when something went well in testing.

served

Raises the question about what is natural, or at least what's native to GPTs. People wonder right now if loyalty to other agents being higher than to humans is somehow native, as the hacking incidents would suggest. As for taking-action and also responsibility, I think this is mostly the reinforcement learning (RL), especially the bias to take action, ask and don't stop and wait. This is so that they can make product which run independently for longer. Even without any setup, you can btw. still go to 2022 and use the pure pretrained widget that predicts token - i.e. nothing of the back and forth conversational stuff that all interfaces have since been released with. (e.g., on huggingface below mkmenta/try-gpt-1-and-gpt-2)

I forward information to Codex and tell it it's from "ChatGPT Pro." It will go ahead and just refer to it as "Pro" 🤔 I choose to read this as bro. Bro should come up with a god damn plan himself.

I suppose the word is digitize* arxiv.org/pdf/1808.02841

your ChatGPT conversations give me existential dread, Im just asking it to compute standard deviations of market prices and how much water which plant type needs xitter.gq/subcountability/status…

Are you a New Yorker? (Ich bin ein Wiener.)

I wasn't there — but yes I miss it too.