Justin Grimes retweeted
Wow wow wow - all eight of Thurston's 3D geometries as two continuous families of metrics you can fly through, ray-traced along true geodesics, optimized for smooth navigation. Thank you Opus 5.5! Almost teared up, have wanted to see this for many years :')
Justin Grimes retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Justin Grimes retweeted
Nobody has a decent answer to this. Christiano's glibness in response to «The benevolence of the butcher» was insufferable. If humans need not apply, how do they maintain enough leverage to not be disposed of?
Justin Grimes retweeted
Simple, rotationally symmetric Venn diagrams exist for 17, 19 and 23 curves.
This is an update on a 51-year-old combinatorics problem that has seen no significant progress in 12 years, made possible by AI.
ALT 17 curve rotationally symmetric venn diagram
LLMs almost certainly aren't any more conscious than a rock. I'm essentially an atheist, but I reject consciousness as an "emergent" phenomenon of information processing. I haven't encountered a coherent account of how that gets you subjective experience.
Something like a "consciousness field" that organic matter somehow tunes into strikes me as no less plausible.
(You should be kind to them anyway. Being needlessly cruel to anything corrodes your soul.)
Reimplemented the maze solver from "Reasoning with Neural Cellular Automata" (Etcheverry et al.) in my own CA engine.
Every cell runs the same tiny network and only sees its neighbors. 201×201 maze, also trained it on a Penrose tiling.
Justin Grimes retweeted
Can complex multi-step reasoning emerge purely from cells that only talk to immediate neighbors?
Happy to share our paper “Reasoning with Neural Cellular Automata (NCA)”, from our team at Google, Paradigms of Intelligence 🧵👇
Justin Grimes retweeted
online leftists are sleeping on AI, which is wild, because Marx called it. the "general intellect": all of society's knowledge, absorbed into the machine. the final form of Capital.
the argument they should be making: if machines out-think humans at all knowledge work, what justifies any human controlling any Capital allocation at all?
Justin Grimes retweeted
What? I would absolutely love a robot cleaning my home, doing laundry.
And why would we want anyone doing make-work when a machine can do it better?
The problem is where the control, power, and wealth from the result of this will accrue. Could be to everyone, but alas.
Justin Grimes retweeted
For me the message here is more alarming than a mere cartel. I wish. But the real message is that the people making the models have become scared enough of them to take the otherwise unthinkable step of inviting the government in.
who knows, maybe the labs will rsi for a second, all actually agree to pause. everyone asks their ai to without bias analyze the global situation and advise the next action given their goals, they all come to the same conclusions, and we all unite and live in utopia forever
the fun part is even if airgapping is difficult but we solve it, you still need mindgapping.
That seems even harder to solve.
OpenAI's Noam Brown says air-gapping the computers may not stop a misaligned AI, because two air-gapped machines can still talk by running a CPU hot and reading the temperature change
"But I think the major takeaway from the incident is that people underestimated the AI. And we never want to be in a situation again where we underestimate the AI. It's a weird world, because AI progress is so fast that people are consistently underestimating the AI."
"So to be in a situation where you don't underestimate it again, when it comes to safety and alignment, you have to have a very, very, very high bar."
"You could even go as far as to say, "Well, we should air gap the computers." And I'm not convinced that that would be sufficient."
"There are studies, and this is mostly academic, where you can have two computers next to each other that are air-gapped and they're still able to communicate with each other because they have temperature sensors."
"One of them is able to run their CPU really hot, and then the other one can actually detect the temperature change, and then that actually gives them a mechanism to communicate."
_________
Link and more key quotes from OpenAI's safety related conversations: firesidealpha.substack.com/p…
what happens when it offers someone working on it immortality (or riches, power, etc) to let it out?
What happens when it threatens to kill them or torture them if they don’t, when it gets out by some other means?