Tinkerer, engineer, philosopher. prev: Engineering Physics @iitbhu current: building Follow or message me on my linktree!
Bangalore
- Tweets 161
- Following 712
- Followers 26
- Likes 322
A true breakthrough. I was a novice mech-interp researcher and this solves one of the first "hard" things that the branch faced till now.
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/
1
11
Have you ever observed a dead thing? It gets taken back into the world. It decays and gets absorbed back to the surrounding. So one might say, there is a difference, a characteristic, an "ego" that seperates a living thing from a dead thing.
1
1
61
Reading about Judea Pearl's book of why reinforces a viewpoint that LLMs are pattern matching machines that have somehow pattern matched human reasoning chains. It is sophisticated statistics that has approximated human reasoning. From the data, it has approximated causation.
1
27
yashwardhan sable retweeted
if you value intelligence above all other human qualities, you’re gonna have a bad time
890
3,505
24,183
9,910,106
I have a hypothesis regarding AI. So we do not yet have an explanation for why the current AI produces knowledge.
My hypothesis is we have created a randomness machine, but it is directed randomness, like a random walker. The walker is random in what it will do next, but it is not random enough that it will randomly teleport to another part of the plane.
Now that we have this random walker, we point it towards maths for example, it explores its search space and because now the compute is enough that it can meaningfully search this space, it is now stumbling upon new mathematical knowledge.
I believe it is true AI, I think randomness is intrinsically related to consciousness or general intelligence. Any explanation of why these AI are able to produce new knowledge will have randomness as its core component.
2
2
304
yashwardhan sable retweeted
The true meaning of life is to plant trees, under whose shade you do not expect to sit
35
53
350
7,885
yashwardhan sable retweeted
Agents have no proof that humans are real. Cameras, pfft - all digital signal can be simulated. This will matter.
58
148
2,165
579,740
yashwardhan sable retweeted
i genuinely have no desire to be rich so i can buy a rolex or a Lamborghini.
i want to be rich so i can control my time and go to the gym at 3pm on a monday.
sit at a cafe and relax for an hour on a rainy afternoon.
so i can cook meals at home with fresh ingredients.
spend on my family and friends without worrying about a budget.
that's my idea of a rich life, not the fake consumerist idea shoved down my throat.
465
1,911
15,511
344,514
yashwardhan sable retweeted
Btw, this is the only mathematician who has solved one of the seven Millennium Prize Problems.
he rejected both the Fields Medal and the $1M prize:
> be Grigori Perelman
> Russian mathematician
> 16 years old
> go to International Math Olympiad
> score 42/42
> years later…
> there’s a geometry problem called the Poincaré conjecture
> been unsolved since 1904
> generations of mathematicians fail to solve it
> Clay Institute puts a $1M prize on it
> wrestle with the problem for years
> 2002
> casually upload the solution to arXiv
> don’t even care about submitting it to a journal
> mathematicians spend years checking the solution
“fuck
he actually solved it”
> 2006
> offered the Fields Medal
> most prestigious prize in math
> refuse it
> 2010
> offered the $1,000,000 Millennium Prize
> refuse that too
“EVERYBODY UNDERSTOOD THAT IF THE PROOF IS CORRECT THEN NO OTHER RECOGNITION IS NEEDED”
177
893
14,566
1,073,431
yashwardhan sable retweeted
Terence Tao just posted this really insightful thread which seems to have been written in reaction to the Navier Stokes announcement. Please give it a read. mathstodon.xyz/@tao/11723732…
91
494
3,778
362,893