Elizabeth Hudson

That much I knew; was thinking more about all the contextual data around the diseased tissue // how much (and profiled in what modalities and dimensions) do you think is necessary to get to the pt where you have a world model that gets you that unlock (or does disease alone do it?)

@ClarkPolner We generate the pretraining data from human tissues.

Massive unlock for biology comes from LLM-powered agents reasoning and running virtual experiments with human bio world models.

Make the Microsoft CEO search for an email on Outlook live on camera

It is with heavy hearts that we announce the deaths of four U.S. Army Reserve Soldiers supporting Operation Epic Fury on March 1st. “We honor our fallen Heroes, who served fearlessly and selflessly in defense of our nation.” said Lt. Gen. Robert Harter, Chief of Army Reserve.

@owl_posting @Dremontjones seconding this question / who tried it and what happened?

@Dremontjones oh neat, where/when else was it tried?

insanely cool ideas for a startup. i dont know anybody involved in these, but it just like such an retrospectively obvious play given the personalized cancer therapy trendlines kernis.health specicare.com

We're mining underpowered relics from the scientific literature when fresh data generation costs pennies. AI co-scientists (like the knowledge-graph tools before them) lean on the academic literature to supply context for whatever the user brings. But what if the literature mostly sucks? How much cheaper and higher-quality does data generation have to get before it makes sense to regenerate the foundations instead of endlessly extracting from them? A typical paper takes a rich, multifaceted dataset → runs a few analyses (only some of the process is documented) → reduces it to a single result → wraps that result in a persuasive narrative (word-count capped, often ~50% marketing). Raw data and full metadata are almost never shared. Even if they were - most of the underlying data are small, skewed, missing metadata, and were generated on old platforms. Negative results were omitted, the research herded around fashionable topics leaving stuff like unsexy benchmarks and context elucidation untouched, and few replications were done because no one gets tenure for being a follower. We still use it because there is zero upfront cost to integrating it. But at what point do those costs exceed the cost of simply regenerating the basics with today’s cheap, high-throughput tools? Without regeneration as the core lever, there will be no science AI worth writing home about.

We need to start talking about legalizing electronic warfare devices under the 2nd amendment. I want radio jammers.

or scenario C with AI (5 shared, 5 unique): in-house: 10 × $20K = $200K/yr platform (20 customers): $5K + $100K = $105K + 30% = $136K saving $64K/yr. is that just "we'll just do it ourselves" territory?

scenario B again (best case platform scenario with 8 shared steps and 2 unique): in-house: 10 × $20K = $200K/yr platform: 8 × $20K / 20 = $8K + 2 × $20K = $48K + 30% margin = $62K platform still wins on ratio. but you're saving $138K instead of $688K. is that worth it?

lots of 'automate data generation' plays in industries that don't have a free internet to scrape for model training (eg bio). some are automating in-house (Lila, Recursion, Insitro). some are building cloud labs as a service for everyone (Emerald, Strateos, Ginkgo?)...but what does an industry have to look like for a third-party cloud lab to beat doing it yourself? napkin math:

@nalidoust How does this relate to Recursion’s approach?

What if we scored drugs not by whether they kill diseased cells, but by whether they push them back to normal? New blog post from Tahoe: we applied cell-state reversal to colorectal cancer using Tahoe-100M, and the results are striking.

@deniswirtz Any chance one of them will be recorded?

I will be giving talks at UC San Diego (Dept Bioengineering) and Sanford Burnham Prebys Institute this coming Friday and the following Tuesday in La Jolla. I will talk about 3D mapping of solid tumors and new CAR T therapies to resolve them. Hope to see you there.

My favorite USA Deep State story is them donating 2 old retired telescopes to NASA and they were more powerful than anything that NASA had then. God only knows what they have now.

The country is waking up to the issue of underprepared college students. The call is coming from inside the elementary school building. This story out of Boston is illustrative: “We do not have the number of students in the entirety of Boston Public Schools performing at grade level in ELA and math to fill the total available seventh-grade seats across the three exam schools.”

This is much worse than just "people are reaching college who can't do basic math" - well, maybe they failed their high school math classes. Oh, no, they got a B in Calculus in high school, and somehow still can't do basic math. That means the high school is lying to people.