this is one of my favorite charts; IEA expert solar forecasts each year vs actual installations people are really bad at wrapping their heads around exponentials

suppose you trained an RL agent to maximize reward across diverse environments then if you dropped it into a new environment, the first question it’d learn to ask is “what is my reward function here?” it might even learn to model the motives of its simulators to figure this out

what do ppl love/hate about ChatGPT memory? in what cases is it useful? in what cases does it fall short?

phrases which take on a different meaning when you work at the AGI lab: “my research was going slowly at first, but now it’s finally starting to show signs of life”

maximizing long-term happiness of the human population is just an applied research problem

AI slop will be looked back upon as growing pains of a premature technology the same way we look at 2023 models that couldn’t do arithmetic bcuz of funky tokenization

ML is an incredibly mimetic field. the narratives that get parroted around why one method worked or another didn’t are often dubious at best. there’s still just so much alpha in actually looking at the data

Emiliano is one of the most cracked people I know, and is working on applying AGI to a giant industry that's still using pen and paper; i think this is a killer opportunity for the right person also a free $10K to whoever can refer someone he hires

emiliano is one of the most cracked people I know and I could not more highly recommend working with him also a free $10K to whoever can refer someone he hires

.@aitechnopagan is an LLM artist/psychologist/jailbreaker who's explored the models deeper than just about anyone i think this type of work is good for the world, someone should fund her!