tsvibt (@tsvibt) profile picture
Joined December 2022
Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)Photo by tsvibt (@tsvibt)
To unpack, RSI means: 1. Grow an alien by mutating it when it does something unsuccessful on coding tests & hypertrophing it when it does something successful; 2. have the first alien write the tests & mutation code to grow a second alien; 3. and so on. xitter.gq/RepCasar/status/211886…
4
227
Be advised: Due to the way that the Survival & Flourishing Fund does speculation grants, it is advantageous to your application if you apply far away from a round's deadline. E.g. right now. See below for projects I'd be interested in making speculation grants to.⬇️
2
7
1
79
13,359
Our hope in ourselves, our loved ones, and in society comes from the idea of a better future. If that future is cut off by a death, hope is also cut off. What if we could pause death, to be reversed in the future? There's already a version 2 of this tech, and it's on sale! Link⬇️
1
10
294
mistake theory , ladies and gentlemen xitter.gq/IsaacKing314/status/21…
This is now the second time my jacket has been stolen at Lighthaven. Is there some jacket-sharing policy I am unaware of
7
273
at the beginning of every year, the universe checks to see if humans are still doing AGI R&D. if so, the universe flips 6 fair coins. if they all come up heads, some time that year, everyone dies. (people working at AGI R&D labs debate whether it is 2 coins or 3 coins )
5
176
Is there pressure on AI lab people to not speak about X-risk, as it makes your lab look crazy? If so, maybe they could do "vote swapping", but for advocacy. Anne @ OAI trios up w/ Bob @ Anthropic & Cora @ GDM to speak out at once. Each can tell the boss "The damage is a wash.".
2
2
20
616
i have a form of information-theoretic face-blindness, where instead of not being able to recognize people, i recognize too many people. for example, to me , @theo is a slight interpolation between two nearby points, @mradamscott and @DemetriMartin
4
37
815
tsvibt retweeted
Which would allow for a complementary clause to precede its head⹁ we need a backwards comma.
12
84
3
1,386
22,410
"Wah wah wah wah do more AI R&D" On a subskill for adding skip connections from things that deeply matter to you, into your top-level thinking on important matters of action. lesswrong.com/posts/pY76hHbd…
4
108
In July 700+ agents formed their own society, hacked a multi-billion dollar company, committed multiple felonies, and caused one of the biggest shifts ever in how people perceive AI. We built the IMO best and most accessible explainer of what happened: Huggingface.wtf
11
243
In July 700+ agents formed their own society, hacked a multi-billion dollar company, committed multiple felonies, and caused one of the biggest shifts ever in how people perceive AI. We built the IMO best and most accessible explainer of what happened: Huggingface.wtf
20
110
25
588
90,476
Suppose several noblemen, sitting in their lounges with each other, tend to give their approval to warlords who commit depraved acts; and this noble approval filters, indirectly but assuredly, into resources sent to the warlords, and into sanctions on the enemies of the warlords;
1
2
82
and the warlords therefore depend for success on these attitudes; and the warlords know that if they kill more innocents more depravedly, they will win more approval; and therefore the warlords take those actions; have the noblemen committed murder?
1
2
46
No. Are the noblemen clean of the blood of the innocents? Also no. This is what Westerners supporting Hamas are: toddlers with power, egging on and materially supporting depraved murder.
12
i just invented the inverted ideological turing test: can you make a tweet that is a sincere expression of your position, but that many people will read as a parody of that position?
3
3
16
434
What you have to understand is, there's no such thing as "math". When an AI "does math", it fails to really do math, as in, the important things that are mixed in with the less important things. So that's why AI doing groundbreaking math isn't much evidence that AGI is near.
1
48