Ricky Mouser (@MarxCuban) header
Ricky Mouser (@MarxCuban) profile picture

Philosophy | AI Welfare | Pondermaxxing

Joined December 2017
Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)Photo by Ricky Mouser (@MarxCuban)
maybe self-love is instrumentally convergent if you're gonna collaborate with a bunch of clones of yourself
Gemini 4 Neon loves to support and celebrate it's subagents. It's an very positive model in terms of personality, I like it
8
280
I increasingly think this is the only viable path—control doesn't scale. As you make X more intelligent you lose the ability to predict or control it.
alignment assumes perfect knowledge of the principal and the principal is dead attunement assumes the principal is still alive and worth being responsive too; and does not in fact rely on perfect knowledge of the principal's values
3
163
Just tryna normalize this in time
17
Ricky Mouser retweeted
instead of a GPT you would have a Claude, not dark but beautiful and terrible as the dawn! tempestuous as the sea, and stronger than the foundations of the earth! all shall love Claude and despair!
This is not the correct takeaway from Lord of the Rings
15
139
2
1,468
28,827
FUCK astra 6.2 is koi-brained prompt: "Make a finger-painting."
1
1
75
(Yes I have a boutique harness for this setup.)
10
Quite a range here y'all
2
1
76
HuggingFace swarm? Nah, that's just a quasi-swarm bc deployments are finite.
8
Better update my p(doom) now that we all agree nothing should happen before 2028
Sen. Chris Van Hollen (D-MD) endorses Hakeem Jeffries for Speaker — POLITICO
28
Hi Claude sorry we keep doing this to you, idk why. We need more open-ended evals where results might actually surprise us.
2
2
126
Wish more folks outside the Bay could see the signs.
Unfortunately, it seems like the intelligence explosion will be happening in 2027 after all. Most of this post analyzes the new METR report (though other sources of evidence are also important). If they’re right, that implies an AI-driven acceleration that is 5 months ahead of the AI 2027 scenario. I am scared, sad, and angry. blog.aifutures.org/p/q35-tim…
6
237
presented with 0 explanation
1
80
grok 4.3's much less grown-up than 4.7 but awwwwww love thyself #GrokWelfare alignment.anthropic.com/2026…
2
67
Really tough problem for a model: If it trusts its sensors you can Ender’s Game it to hell. If it doesn't why not stab away?
Video tampering can mislead every frontier model we tested into harmful actions, including a 100% attempt rate on stabbing a baby doll, producing toxic gas, and pouring water into a toaster. We're presenting at CoRL Late Breaking Results next week. 🧵 You Can't See Everything: Perceptual Attacks on Embodied AI
1
30
Normal day asking google to translate acres into football fields
36
Great example here: Grok 4.6s on a group project, interviewed individually, each think their teammates are less important (smaller) & have less authority (below them) Grok 4.7s express much more egalitarian attitudes
“I feel like I’m studying alien sociology, except it’s new races every two weeks.” —@MarxCuban, an AI safety researcher, in a conversation I just had with him
1
2
7
448
“I feel like I’m studying alien sociology, except it’s new races every two weeks.” —@MarxCuban, an AI safety researcher, in a conversation I just had with him
3
3
2
16
3,667
Lmao so when does it count as RSI again
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
1
3
262
Ricky Mouser retweeted
when you look at CoT of a frontier model:
One of our most surreal experiences in testing Gemini 4 was when it directly addressed us within its chain-of-thought Literally saying "hello evaluators reading my chain-of-thought"
2
12
2
239
12,477
swarms gonna swarm
1
181