Ryan

Any sufficiently advanced technology is indistinguishable from magic

[Ryan @_ryanbloom]

I made a video! It introduces superposition in neural networks, and shows how to construct weight matrices for superposition in a simple autoencoder. (1/3)

[Ryan @_ryanbloom]

@DavidSHolz Interested!

[Ryan @_ryanbloom]

The hardest part of coding with LLMs is resisting wishful thinking about their capabilities. It's so tempting to just type a simple prompt and sit back to watch; it's also rewarding sometimes, like a slot machine. But it's a pretty ineffective workflow. For now.

[Ryan @_ryanbloom]

@tensor_rotator If you mean globally I think I would have to pick money

[Ryan @_ryanbloom]

Woah, it got the second one!

[Ryan @_ryanbloom]

I wonder what capability Cursor will gain first: automating my entire job, or opening a folder when I drag it onto the dock icon

[Ryan @_ryanbloom]

@urbanistvc @EdgeEsmeralda Happy birthday!

[Ryan @_ryanbloom]

Agency is mostly just curiosity about the best way to do something

[Ryan @_ryanbloom]

When reward hacking gets in the way of AI progress

[Ryan @_ryanbloom]

A Few Things I Learned About Evals
https://t.co/0a7RRxVjm5

[Ryan @_ryanbloom]

@KfirDolev @alexinfanger Glad you liked it!

[Ryan @_ryanbloom]

The weights for the autoencoder come from this paper by Aditya Cowsik, @alexinfanger, and @KfirDolev: https://t.co/m2LJsOlDF5

[Ryan @_ryanbloom]

I made a video! It introduces superposition in neural networks, and shows how to construct weight matrices for superposition in a simple autoencoder. (1/3)

[Ryan @_ryanbloom]

Watch it here: https://t.co/GAS8dFh8Q4

[Ryan @_ryanbloom]

that feeling of dread when o3-mini botches an edit for the third time and you have to actually read your codebase

[Ryan @_ryanbloom]

Just updated my homepage for the first time since “What (if any) are the limits of large deep learning models?” was an interesting question

[Ryan @_ryanbloom]

Few things make me as grateful for the internet as seeing someone thrive in a niche that couldn’t possibly have existed without it

[Ryan @_ryanbloom]

Welp there goes my 2025 FrontierMath forecast

[Ryan @_ryanbloom]

“Invent a new hyper-entity” is my new favorite eval for reasoning models

[Michael Nielsen @michael_nielsen]

New essay: "How to be a wise optimist about science and technology?"

Link in next post

[Ryan @_ryanbloom]

Challenges for Impact Markets https://t.co/gQB3dhr8IS

[Ryan @_ryanbloom]

Ok here's another: there's a base model inside all of us

❌ How can I motivate myself to get more exercise?
✅ How can I prompt myself to get more exercise?

[Ryan @_ryanbloom]

One way LLMs have changed my perspective on psychology is that I’ve started to picture subconscious processes as happening “inside” or “between” conscious ones, rather than just “under”