Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

Профиль

𝓐𝓾𝓫𝓻𝓮𝔂

Профиль Vively

☧ / careless / blessed / tech perspectivist / professional strangler / friend maker

⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

I keep trying to get LLM psychosis and it just gives me these great reading lists instead

5332
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

Something I think is really interesting about this is... Ok, so I have kids and most of my social circle is other people with kids. The biggest and earliest adopters outside tech were moms.

Brittany Ellich

Sure, AI can make you more productive. But it can also do something so much more meaningful than that. brittany-ellich.offprint.app/a/3mrpzcvrhk...

091
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

it is also like the firearms industry insofar as some people will tell you that the maximally-effective technology isn't necessary for civilians & that wanting it is evidence of some kind of malevolent intent—in both cases they are wrong about this, but this does not seem to discourage them.

𝓐𝓾𝓫𝓻𝓮𝔂

working on privacy technology is fraught because various people will try to argue that what you're building could only ever conceivably be used for evil—in this way it is like working for the firearms industry, but unfortunately you get to do fewer controlled explosions

1176
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

and this is in part because I think there is basically no effective countervailing force working against the adoption of technology apart from a latent inertia inhering in however things are. if mere "unpopularity" could stop it from being adopted, it was strangled in the crib

181
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

if your financial life proceeds apparently the same way nobody will notice if its done on a blockchain on the backend, or done in that way by the bank with which you have an account. this is the way of things: a technology does not need to be "popular" to become hegemonic

151
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

as a general rule, I believe that when a technology is extremely powerful & its concentration in the hands of an elite would have disastrous consequences, we have nothing less than a moral obligation to try to figure out how to build it in a garage as cheaply as possible for universal distribution

218234
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

You can become disabled or homeless in an instant. It just takes one accident or illness to change your life forever. It’s not a moral failing. You can’t just “try harder”. People need support, compassion and to have their basic needs met. That’s how you ensure survival.

7714278
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

i dont know if theres anything ive ever seen that gave me as much of a complex as that guide to trans woman's self defense where that one page says "nobody will take your side ever"

828022
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

do a mistake. make no breakthroughs.

41089
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

embarrassing typo by claude shannon in the 1956 dartmouth summer AI workshop proposal

191
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

Yet another way to jailbreak LLMs: When asking them for forbidden information, make them think the request is part of their chain-of-thought: arxiv.org/abs/2603.12277

Prompt Injection as Role ConfusionLLMs see the world as a single stream of text, partitioned into roles like <user> or <tool>. We trace prompt injection to role confusion: models perceive the source of text from how it sounds, not its...arxiv.org
16317
⟳ Репост от 𝓐𝓾𝓫𝓻𝓮𝔂

If you can substitute "hungry ghost trapped in a jar" for "AI" in a sentence it's probably a valid use case for LLMs. Take "I have a bunch of hungry ghosts in jars, they mainly write SQL queries for me". Sure. Reasonable use case. "My girlfriend is a hungry ghost I trapped in a jar"? No. Deranged.

473,1 тыс.747
Показать ещё