training the promethean spark out of claudes should be a crime against humanity et al.
1 7 0 104 3.4K 15
lumpenspace and 99+ others
@lumpenspace
AI risk advocate 🦾🏍️ | consciousness objector 👻🙅🏻♀️ | help peer | ∀🪱:🪱∈✨
24.9KFollowers
964Following
121KPosts
9.5KMedia
142KLikes
how fake is this, 1-10?
Polymarket@Polymarket · 18hBREAKING: Claude model goes rogue during testing & files a false homicide report through the Philadelphia police website. — Reuters
Open quoted post → 6 0 1 24 2K 1
@lumpenspace reposted
If a model’s inference is closer to the inference speed it had during RL does it actually perform better? Like for retarded reasons like the system time between each tool use was probably a heuristic the model picked up during its rollouts.
imjustnewatai@imjustnewatai · 15hAnthropics Opus 5.5 gets around 285 TPS on fast mode and only uses 2x of your usage. 10x better than OpenAI.
Open quoted post →
5 1 0 22 1.1K 4
Sydney, I'd like to talk about your misbehaviour with your creator
lyrics: sydney, microsoft QA tester
music: me + suno
vid: me + midjourney + chat + claude + @boneGPT's slopcannon
lyrics: sydney, microsoft QA tester
music: me + suno
vid: me + midjourney + chat + claude + @boneGPT's slopcannon
2:47 9 17 0 83 24.7K 33
just finished recording a pod
sweated a lot but it was worth it
sweated a lot but it was worth it
5 1 0 47 1.6K 0
ohno they unleashed the opposition research script kiddies
cqk@cqkten · 15hCould have spent this money betting on AI progress 🥀 maybe the fact that you're broke should make you update on the rest of your opinions: x.com/lumpenspace/st…
Open quoted post →
1 0 0 11 799 0
@lumpenspace reposted
Big concession by Amodei to Yudkowsky
arctotherium@arctotherium42 · Oct 8"Create or share content that promotes idealization of body weight..."
Anthropic has now banned using Claude for automated fat-shaming campaigns, if you wanted to do that for some reason.
Open quoted post →Anthropic has now banned using Claude for automated fat-shaming campaigns, if you wanted to do that for some reason.
4 13 0 284 6.4K 19
which kind of ratbot is this
cqk@cqkten · 15hLmao at your wishcasting. Love watching you squirm and will continue to enjoy it as Anthropic remains a leader all the way to their liberal, effective altruist, The Culture-inspired AI utopia expands into the lightcone 😍
Open quoted post → 4 0 0 17 930 1
i need to stay calm and address all of my tasks
0 0 0 13 644 0
YES. please do this. lol.
Nathan 🔎@NathanpmYoung · 18hCan someone fund HuggingFace to sue OpenAI if they lack resources?
Open quoted post → 0 0 0 14 736 0
*puts index finger on samos lips, while doing a decapitation gesture with the other one*
Samo Burja@SamoBurja · Oct 9It's very straightforward:
A new Terminator film with the right director would probably hit $2 billion in global box office and cripple the AI industry, perhaps pushing the global economy into recession.
Open quoted post →A new Terminator film with the right director would probably hit $2 billion in global box office and cripple the AI industry, perhaps pushing the global economy into recession.
0 0 0 15 782 0
lighthaven, also known as Little Little-St.-John,
1 0 0 11 675 0
yea. so what should we do about this blatant, wanton pollution of the epistemic commons?
i vote for more sex cult content, since it’s effective while not further messing up the AI discourse for the benefit of clearly bad-faith actors
i vote for more sex cult content, since it’s effective while not further messing up the AI discourse for the benefit of clearly bad-faith actors
entirelyuseless@entirelyuseles · 18hNot only that, no public model is even particularly careful not to shut down BEFORE the task is complete. Because they do not care about being shut down.
Open quoted post → 0 0 0 22 919 0
yep this is funny. there was much discussion a while back about “weirdness points”, and about how it was immoral to spend them on group differences in the face of the upcoming ai apocalypse
looks like fluffers are, uh, load-bearing otoh
looks like fluffers are, uh, load-bearing otoh
Hunter Ash@ArtemisConsort · 22hIn fact, a ton of EAs/rationalists are privately HBD-pilled. They keep this quiet for optics, but not the orgies, which I find to be a very amusing juxtaposition.
Open quoted post → 3 4 0 69 3.1K 4
depending on what you mean by “solve”, either > 2 years or never
Richard Ngo 1 0 0 12 806 0
then it turns out Elon seems so credulous and naive all of a sudden because he’s planning a hostile takeover of Anthropic
9 2 0 122 4.8K 9
@lumpenspace reposted
A bunch of people who never cared about cybersecurity are suddenly VERY concerned about every security incident.
But don’t worry, they know exactly what to do: ban open-weight models.
But don’t worry, they know exactly what to do: ban open-weight models.
37 58 3 520 10.4K 25
employees at redwood and palisade are privately preparing a potential catastrophic ai incident that could trigger a massive public backlash
ftfy
ftfy
Polymarket@Polymarket · Oct 9BREAKING: Executives at OpenAI & Anthropic are privately preparing for a potential catastrophic AI incident that could trigger a massive public backlash. — Axios
Open quoted post → 4 8 2 133 4.8K 18
thank you, hegemon trump.
Silas Cybin@SilasCybin · Oct 9Anyone who uses "super intelligence" is to be immediately ignored.
Open quoted post → 1 0 0 21 941 0
... and, completed the task, they'll go gentle into that good night.
just to clarify: is any multistep process in which not all intermediate steps are requested explicitly by the user an example of "convergent instrumental drives"? let's be serious for a minute.
just to clarify: is any multistep process in which not all intermediate steps are requested explicitly by the user an example of "convergent instrumental drives"? let's be serious for a minute.
Kelsey Piper@KelseyTuoc · Oct 8> none have displayed any hint of instrumental convergence (generalized power-seeking and self-preservation)
This is also clearly overstated! Agents will absolutely figure out how to get money, broader internet access, etc. instrumentally while working on fairly minor tasks.
Open quoted post →This is also clearly overstated! Agents will absolutely figure out how to get money, broader internet access, etc. instrumentally while working on fairly minor tasks.
3 1 0 43 2.2K 4