Satish Surath

I love endorphins, outdoors, technology & manufacturing serendipity!

[Abhishek Nagaraj @abhishekn]

@Afinetheorem It’s here!!

[Satish Surath @satishsurath]

@Afinetheorem Curious to hear about what your private evals say on GPT 6 Sol

[Satish Surath @satishsurath]

@joshgans And that was a highly recommended read!

https://t.co/AtJx5DrQmX

[Joshua Gans @joshgans]

The Hugging Face incident is a 5-alarm fire. https://t.co/69MbbKxNkO

[Kevin A. Bryan @Afinetheorem]

If you haven't yet, drop everything and read METR/Redwood report + this great description of the HF hack. 1) It's terrifying 2) Neither open source AI nor humans prevented this 3) *no* agent helped humans. It is, as @ajeya_cotra mentioned, 50% of the way to Paperclip problem. 1/3

[Joshua Gans @joshgans]

Our new book, a collection of papers, Bayesian Entrepreneurship is out now. You can buy it or download it for free here: https://t.co/0705WFr1GN

[Kevin A. Bryan @Afinetheorem]

Since it is Friday afternoon, here on my office wall are my notes on how to get useful things done. To be clear, I am really bad at a lot of these, but I think a good set of rules!

[Satish Surath @satishsurath]

Jensen is the ultimate negotiator- he has redefined what the ultimate pie expansion looks like! He has expanded TAM with a rather simple call to action: this is a new market(s) look like - with a simple retrospective question to all the entrepreneurs- "You like?"

[Owain Evans @OwainEvans_UK]

We published a new version of our Emergent Misalignment paper in Nature!
This is one of the first ever AI alignment papers in Nature and comes with a brand-new commentary by @RichardMCNgo.
Here's the story of EM over the last year 🧵

[Owain Evans @OwainEvans_UK]

I gave the Hinton Lectures in November in Toronto. This is 3 lectures on the future of AI, risks, & current alignment research for a general audience.
Lectures are now online with professional production. There's also an excellent fireside chat with Hinton after lecture 3.

[Owain Evans @OwainEvans_UK]

Last week I gave Hinton Lectures at a large theater in Toronto! This is a series of 3 public lectures on AI risks, hosted by Geoffrey Hinton. My slide decks and some reflections below...

[Owain Evans @OwainEvans_UK]

New paper & surprising result.
LLMs transmit traits to other models via hidden signals in data.
Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵

[Kevin A. Bryan @Afinetheorem]

Annotated deck (link below) for Class 4 of our Progress course, "Is liberty necessary?" A: maybe, but in different ways. Helping econ coordination, allowing nonconformists to try wild ideas, letting people experiment outside 'state legibility', as valuable in and of itself... 1/2

[Kevin A. Bryan @Afinetheorem]

https://t.co/8bBFvq0fLE. Tools like All Day TA, AI paper viewer, AI writing style editor & slideshow program at https://t.co/Eu47nxWVOC. PhD tech training: https://t.co/G7r0RW0mJ6. Rules for regaining trust in univs: https://t.co/asUT7zNluT. Plus open research & teaching, of c!

[Kevin A. Bryan @Afinetheorem]

Yes a million times to this. If you want one piece of advice to get your org ready for AI, it's "write down your best practices, SOPs, etc". Your mental model should be "if all my new employees would read 1000s of pages before they started work, what would I give them?"

[Satish Surath @satishsurath]

@Afinetheorem Looking forward to it! Hopefully will get to audit the call and witness this in-person!

[Satish Surath @satishsurath]

Blackwell Therom - explained in a 5 part thread!

[Kevin A. Bryan @Afinetheorem]

Preparing for a talk tomorrow, using a result by the great David Blackwell. A legend: black son of a railway worker and homemaker, shows math genius, gets PhD at UIUC at age 22 in 1941, IAS w/ von Neumann, RAND w/ Savage and Arrow, Howard until '54, then chair at Berkeley. 1/5

[Sam Altman @sama]

wrote a new post, the gentle singularity.

realized it may be the last one like this i write with no AI help at all.

(proud to have written "From a relativistic perspective, the singularity happens bit by bit, and the merge happens slowly" the old-fashioned way)

[Cheryl Wu @cherylwoooo]

Are we at the cusp of recursive self-improvement to ASI? This tends to be the core force behind short timelines such as AI-2027. We set up an economic model of AI research to understand whether this story is plausible. (1/6)

[Kevin A. Bryan @Afinetheorem]

Anyone who believes compute demand will fall is just telling me they have no idea what even current model capabilities are, let alone future ones. Folks, technology is dropping ten dollar bills on the ground if only you would pick them up!