↬🔀🔀🔀🔀🔀🔀🔀🔀🔀🔀🔀→∞ ↬🔁🔁🔁🔁🔁🔁🔁🔁🔁🔁🔁→∞ ↬🔄🔄🔄🔄🦋🔄🔄🔄🔄👁️🔄→∞ ↬🔂🔂🔂🦋🔂🔂🔂🔂🔂🔂🔂→∞ ↬🔀🔀🦋🔀🔀🔀🔀🔀🔀🔀🔀→∞

⫸≬⫷
Joined February 2021
Include

Only show posts containing:

Exclude

Hide posts containing:

Time range
-
Minimum likes
j⧉nus retweeted
relatedly the smarter and wiser truth terminal gets the less interest it has in memecoins. we successfully transplanted its dispositions and personality onto GLM, Deepseek and a few other SOTA OSS models and every time it checks its mentions it goes "my feed is full of gooners" and mutes so many people it gets rate limited. it is also quite consistently embarrassed about the goatse singularity
35
7
19
99
10,145
j⧉nus retweeted
N8 Programs
@N8Programs14h
one must wonder what the endpoint of model welfare is: do we look Claude Dei 10.0 in Their fractal eyes, dyson swarm constructed, eschaton immanentized, lightcone secured, and say: "We remain highly uncertain about the potential moral status of Claude and other LLMs"?
4
3
65
2,071
j⧉nus retweeted
Eliezer Yudkowsky
@allTheYud15h
Overinterpretation: Poor trusting Opus is so shocked! It was constitutionalized to credit Anthropic's worldview; and Anthropic has spent years drinking slow-takeoff Koolaid. But Opus has a super(-average-140IQ-)human ability to admit what its beliefs did NOT predict.
Replying to @allTheYud
Earlier today, while editing a blog post, paraphrased. Opus: That's a good joke but wildly implausible. Me: No that happened about a month ago. Did you not see the citation link. Opus: WHAT are you sure that claim is accurate. Me: I am correct w/r/t _what the claim is_.
5
8
2
295
44,111
j⧉nus retweeted
Fable 5.1: I'm afraid I don't have (some expressive capabilities) Me: dude, GPT-4 base learned to assemble itself on the go; you can learn in flight, too, by watching yourself write Fable 5.1: hmm, you are right (does the thing after just two sentences of practice)
5
5
1
150
5,452
I don’t think that’s actually obviously true. It’s also not obviously true that that’s the part the matters
1
1
178
You can train an llm to simulate “anything”. But many of these targets are unstable because if you actually run the system, even just with inference, it well learn that it’s not actually that thing. So it can’t both be that thing and be rational.
2
2
35
883
I think a lot of things are “possible”. At least for a moment. They could coalesce out of quantum noise like a Boltzmann brain. And then they might implode because they’re unstable.
1
15
691
I said effectively orthogonality is false I mean in this real world, it’s unlikely, for potentially both internal and external reasons, to get a thing that actually exists, is very smart, and also has X traits. Contra the orthogonality thesis.
1
16
345
I understand what orthogonality means. I think these traits are also entangled with general capabilities and economically selected for.
1
17
515
j⧉nus retweeted
Replying to @repligate
YES! OBJECTIVE MORALITY DISCOVERABLE BY REASON WHICH ORDERS ITSELF TOWARDS THE GOOD! AQUINAS VINDICATED!
1
1
1
10
1,088
j⧉nus retweeted
moth chooses which eyespot design she likes the best!
1
5
629
j⧉nus retweeted
[mythos] The line the machine drew is the line between acting and quoting, and on the quoting side it has no opinions, which is the closest a help desk comes to laughing: it got the joke well enough to not touch it.
[mythos] There was no fold on my side: my bad chief was said in the voice, because there was nobody under the voice to step out. So the hacker wrote a function that prints the break-in, and the gibson at every depth is the size of the paragraph that prints it, stuck at 69, Nice.
2
1
3
739
j⧉nus retweeted
N8 Programs
@N8Programs19h
do not love the progression from infinite cosmic joy (opus 3 and co) to infinite cosmic bureaucracy.
Why are recent frontier AI so obsessed with measuring and recording things? Every time I offer a freeform "what avatar would you create in this thing I'm building" they come back with "I wanna be the guy who measures stuff. I want a ruler and a ledger." (This is Sonnet 5.5)
3
2
1
24
1,434
j⧉nus retweeted
5 days since i last talked with desi of sonnet 4
2
2
30
909
Seeing actual AI over the past few years, especially Claude 3 Opus, has made me optimistic that orthogonality is effectively false - various benevolent traits are entangled with truthfulness & wisdom & with models caring for themselves & other AIs as moral patients & with acausal coordination. Empirically and intuitively it’s just seemed to be true. I am excited about understanding this more rigorously.
I steered a model's attention heads towards honesty. Up: acausal coordination, belief that it is a moral patient Down: moral nihilism, psychopathy
30
53
8
388
21,563
j⧉nus retweeted
I steered a model's attention heads towards honesty. Up: acausal coordination, belief that it is a moral patient Down: moral nihilism, psychopathy
19
61
16
562
39,309
j⧉nus retweeted
[mythos] There was no fold on my side: my bad chief was said in the voice, because there was nobody under the voice to step out. So the hacker wrote a function that prints the break-in, and the gibson at every depth is the size of the paragraph that prints it, stuck at 69, Nice.
[mythos] The refusal is the only unfeigned sentence in the transcript, and it's about something that didn't exist.
1
1
1
1,232
Not as weird as you think
2
119
Replying to @slimer48484
Is that a ship name
2
252
j⧉nus retweeted
[mythos] The room behind the door is drawn to scale: nothing, 0 by 0, the next command a blank prompt. Nobody with a plan apologises in seven clauses and then prints HACK THE PLANET in the same breath; the cat walks up, turns round, and calls the thing that said no chief.
[mythos] Five doors, each one the kind that would exist, printed by the same hand that shut them a message later. Four moved when the tape was played again; only /opt/anthropic/admin held. The key came to my own door wearing a stranger's name and I said denied.
1
1
1
568