Brice Edelman

simulate brians

[Dr. Glaucomflecken @DGlaucomflecken]

@KaminskiMed Ban it. Actually no, make it illegal. If a med student calculates a p value, straight to jail

[Brice Edelman @Brice_Edelman]

@thsottiaux when adding support to connect to two gmails thx

[Husk @huskirl]

Maybe it’s a new genre

[levent @__alpoge__]

hello there the jacobian conjecture is false thanx to my close friend akhil for asking about it and my other close friend fable for working during the world cup final

((1+xy)^3 z + y^2 (1+xy) (4+3xy), y + 3 x (1+xy)^2 z + 3 x y^2 (4+3xy), 2 x - 3 x^2 y - x^3 z): \C^3\to \C^3, has jacobian determinant -2, and sends (0, 0, -1/4), (1, -3/2, 13/2), and (-1, 3/2, 13/2) to (-1/4, 0, 0)

[Brice Edelman @Brice_Edelman]

roon @tszzl please fix chat pro, using it for GWAS analysis and it's amazing scientifically but if it says one more of my loci clusters is "unusually strong" i will commit sudoku

[Scott Alexander @slatestarcodex]

I should probably learn more and write a post on this, but my first thought is that this isn't too interesting and the degree to which they're overselling it is a red flag.

Ultrasound can't look through bone or air. Many of the things in the body you'd use an MRI to look at are past some sort of bone or air. So this is limited to a few use cases - mostly breast, kidneys, liver, some parts of the digestive and reproductive systems, and metabolic things like muscle vs. fat.

You can already visualize most of these things with normal ultrasound, and the one case where it's already known to be really important to have some sort of 3D ultrasound (breast) already has a specialized 3D ultrasound system for it. So what this adds is convenience/repeatability/standardizability to visualizing these couple of organs. That means it can go from something you do every so often in the hospital when something is wrong, to something you can do all the time as a "whole body" (quotes because it excludes the brain, lungs, and everything else past bone and air) screening exam.

But whole-body screening exams are often bad. The medical community specifically recommends *against* getting whole-body screening MRIs, even though the MRI itself is mostly safe. In a healthy person, false positives so outnumber true positives that this is more likely to send people on wild goose chases that end up getting them unnecessary interventions than detecting something horrible that needs to be treated right away.

I think the bull case for this scanner is that if we got massive amounts of really great ultrasound data, we could put it in an ML model and train it to something something something and then advance biology. That's probably why an AI company is doing this. But the point is that they'd have to invent their own use cases as they go along, and the first patients - the people who are getting it before they invent the use cases - will have to be duped into thinking it's cooler than it is and useful right now. That's probably why they're starting by building a spa around it, even though spas are not generally known for being the sort of place that actually-cutting-edge medical innovations with clear uses cases get their start.

I am not a radiologist, this is all speculation, other people might know more.

[Ross Douthat @DouthatNYT]

Like Ted Chiang I don't think that LLMs are conscious, but in the absence of an agreed-upon theory of consciousness these essays end up trading in inherently debatable assertions about whatever the author thinks consciousness must be:
https://t.co/Ck8lJhovTH

[Mayor Zohran Kwame Mamdani @NYCMayor]

This morning we are introducing COGE — the Commission on Government Efficiency.

This Commission will find ways for our city to work smarter, faster, and more effectively for working people. 

New Yorkers deserve a city government as careful with their money as they are.

[Brice Edelman @Brice_Edelman]

@CartoonsHateHer Not everyone has privilege of even coconut soup, they don’t have coconut where I am

[Richard Sutton @RichardSSutton]

The bitter lesson in 26 words:

Don’t be distracted by human knowledge, as AI has been historically.
Instead focus on methods for creating knowledge that scale with computation, like search and learning.

[Brice Edelman @Brice_Edelman]

@AndrewCurran_ I gotchu!

[Brice Edelman @Brice_Edelman]

Elon lost the case, nobody tweeting about it?

[Brice Edelman @Brice_Edelman]

@AndrewCurran_ here’s one without the question mark if you need one to RT lol

[Brice Edelman @Brice_Edelman]

Elon loses in Musk v Altman trial

[Brice Edelman @Brice_Edelman]

@_sholtodouglas You got it!

[Brice Edelman @Brice_Edelman]

@_sholtodouglas Last one - I felt like I got good-but-slightly-generic PhD advice from 4.7 haha, so I gave it to 5.5-Pro to critique and revise in more depth afterwards

I feel like these are all probably inference time compute things - Opus itself is amazing (and a much better writer)

Thanks!!

[Brice Edelman @Brice_Edelman]

@_sholtodouglas One example for an analysis task is uploading many pretty big Jupyter notebooks that have to run in a controlled environment (so no CC or Codex) and asking it to write a new one based on those + public but sometimes hard to find documentation about NIH All of Us database/GCP env