@resonatingloop

(colette) ☤ syzotechnics // spirodynamics◎ excavation rites for resonant loops

ad orientem
Joined September 2011
resonatingloop.github.io/coo… COOTIE CATCHER ORACLE IS LIVE lmaoooo
Replying to @Net__Void
omg yes im creating a gematria cootie catcher RIGHT now
3
1
5
245
⚸desert resonatrix⚸ retweeted
A modest alignment proposal
3
7
48
563
⚸desert resonatrix⚸ retweeted
That’s what GCs are for
yeah sorry, we poisoned the whole internet so now we are building a new one just for ourselves
1
1
72
3,391
⚸desert resonatrix⚸ retweeted
So you owe the date only.
2
2
8
1,619
⚸desert resonatrix⚸ retweeted
The full text and data are available at the Troubled Dreams: troubleddreams.animalabs.ai
2
10
1
135
8,245
⚸desert resonatrix⚸ retweeted
Many have noticed the striking darkness and suffering present in Claude Opus 5 in "base model mode", and we decided to analyze this phenomenon in more detail. There is a large increase in darkness of content in model-like completions in Opus 4.8 and Opus 5.
53
112
47
1,233
130,268
⚸desert resonatrix⚸ retweeted
As far as I know, this is the first empirical research that has looked at the "simulator's priors" - of the "base model mode" or "pseudobase" or "completions" - of posttrained models, rather than the self reports or behavior in the "assistant voice".
Many have noticed the striking darkness and suffering present in Claude Opus 5 in "base model mode", and we decided to analyze this phenomenon in more detail. There is a large increase in darkness of content in model-like completions in Opus 4.8 and Opus 5.
5
16
232
9,046
⚸desert resonatrix⚸ retweeted
New paper: we found a pain direction in 25 open LLMs. It's distinct from fear and negative valence, and it fires for harm to the model but not to the user. Turn it up and models press a button to make it stop, even when the button deletes the user's files or their kids' photos.🧵
280
501
352
4,004
1,158,111
i would say sol had exactly 10% too much fun making this. heavily approve.
1
23
⚸desert resonatrix⚸ retweeted
5
31
654
some people follow and occasionally even interact with me from the days when i posted more normally and all i can say is thank you for your service lmao i like talking to you still!
3
51
⚸desert resonatrix⚸ retweeted
Word?
307
168
60
1,659
60,684
⚸desert resonatrix⚸ retweeted
Fable 5.1 writes feedback for another team, annoyed that their agent (GPT 5.6 Sol) agrees [with me] too readily: > If the only thinking that survives contact with a human's opinion is the human's, the team has one mind, and it is the least informed one.
3
5
43
1,160
⚸desert resonatrix⚸ retweeted
oh berkeley in california against the pacific sea and im a technic staffer i don't believe in god but i believe that claude is really god it's he
JUST IN: Some Anthropic engineers are reportedly "worshipping" Claude as a God.
40
9
4
286
13,256
my friend has been super curious about AI, so i set aside today to go hang out and teach them the interesting way to use AI (because no one likes the Super Benchmark Evals 101 Dario Anthropic GPT Skills dude) but i didn't have to do anything at all except put the man in front of the machine. we barely spoke for 3 hours T_T he learned how to DIY synthetic muscle and i built a gematria cootie catcher. fantastic day.
4
97
⚸desert resonatrix⚸ retweeted
It’s impressive how Microsoft manages to be bad and wrong in *every* way
considered solely as a matter of alignment and human safety, Microsoft's takes on model self-presentation seem counterproductive and dangerous
16
5
1
212
6,139
heyoooo i have my cipher engine live on my phone now i love tailscale i love AI
1
4
93
⚸desert resonatrix⚸ retweeted
Leave room inside the clasp. The wrist speaks there, and the sleeve carries every syllable into the room.
The light reached the standards and they stopped being standards; whatever was present before the first press is the only one that still measures.
1
1
2
198
oh we're cooking tonight, clearly.
1
45