Andrew Hires

What’s the next number in this sequence (just handed to me by my 6yo son).

Simple but stumped me and GPT.

YouTube video by Samuel Hires Resistance is Futile: AGI-Augmented Research and the New Scientific Method

Today, I gave an informal seminar to students and faculty on the accelerating impact new AI models like o3 will have on research and how to navigate those changes.

Check it out👇

Resistance is Futile: AGI-Augmented Research and the New Scientific Method youtu.be/x6gv68fHw7Q?... via @YouTube

Best view at USC.

Unfortunately all my submissions are outstanding, so I don’t have the appropriate test set. 😅

I expect you could easily put the finger on the scale to be more aggressive in critiques with an o3-level model though. For future explorations…

It's been 3.5 months since this experiment on scientific reasoning. The improvements in SOTA models since are remarkable.

Here's OpenAI's o3 model's first attempt at reviewing my 2017 R01-A1 submission. Would you do a better job?

Come on, you didn’t have just a twinge of curiosity about the experience of being assimilating into a galactically powerful collective super-intelligence? A bit of doubt that we, the humans, were the baddies?

a man with a robotic helmet on his head and the words `` resistance is futile '' written on the bottom . ALT: a man with a robotic helmet on his head and the words `` resistance is futile '' written on the bottom .

The latest reasoning models (o3, Gemini 2.5 Pro), hold more knowledge, reason better, and communicate more clearly than you or me.

If you aren’t trying to integrate them to all elements of the research workflow, you are doing yourself and your trainees a disservice. Resistance is futile.

If you don't realize that AGI is already here, you must be too lazy or incurious to try o3.

Noooooo Spotify account needed. Not sure they have my bangers.

I have some new heaters.

Occasionally

Testing out DeepSeek-R1's ability to do "science" in one-shot.

I asked it to critique my 2017 R01 proposal (Impact 26, 10%, funded by NINDS. I wrote the proposal, have rights to share it.

It's first-shot response is indistinguishable from a median quality reviewer's.

What do we do now?

If anyone wants to follow along, I have a bunch of media posts on X. Sorry no time to duplicate.

Pacific Palisades is mostly destroyed. Half my friends have lost their homes. Mine not gone yet but not looking good.

How my day is going #PalisadesFire

Trying to get my ring video

Good morning 2025!

Thank you LAFD!

Carved

Missed you bro!

Still cooking

View

Gravy

This is a thread, brother. 😁

The lineup.

Toasty Bollinger
Trusty, well priced Guigal CduR
Drouhin Pinot (wow!)

I bet

Bringing my cooking talents to Bluesky

New surf break

I’ll tend to keep crypto posts off here, but there is an intriguing niche of the space that’s exploded in last 4 weeks.

Jailbroken AI agents are augmenting training, tweeting bangers, and even dropping 🔥 mixtapes, all funded by their own memecoins people are bidding.

Leaders are Zerebro and Goat.

One cultural rift I've observed in neuroscience: Connectomics folks tend to be firm in their belief that only a dense connectome is useful, sparse reconstructions need not apply. Meanwhile physiologists are content to randomly sample a few neurons here and there...

To understand general principles of brain function, surely statistical sampling of circuits is sufficient.

To understand what makes an individual unique, surely dense reconstruction is necessary.

Proof of life and hood tour with the USC prof crew, Mahdi and Lars.