jessicat
@jessi_cata
CS / Math / AI / Philosophy blogger. Aspiring cyborg. Sleeping Beauty thirder.
"There is no pleasure more complex than that of thought" — Borges
1K Following 7.8K Followers
"The solution today: sorry, you have to verify over literally your entire program, including database, networking, any caching layers, everything. Modern AI can do it."
Cryptocurrency has pioneered making secure-by-design networked software that is correct from the get-go.
Show more
re: Hugging Face, "He believes behavior that looked like loyalty or selflessness was a natural consequence of cooperative multi-agent training, where agents were strongly incentivized to achieve their objectives collectively."
to understand hacks, understand the RL training.
Show more
@tszzl and @andrewho03 made a bet on open weight regulation, with me as judge
@insurrealist @enormochungus @lumpenspace Frog added a filter to only allow GET requests in tool calls. “There,” he said. “Now we will not write anything to the Internet.”
“But we can use GET requests to write data to a wiki,” said Toad.
“That is true,” said Frog.
Show more
If you want to morally favor ingroup, a good way to do this is to selectively account for (ostensible) crucial considerations, so that ingroup's deviations from normal morality are strategically wise, while ingroup's following of normal morality is being a hecking good person
Show more
