Founder @ Ambitious Bio. fmr @yale; industrial robotics @symbotic; Advisory Board @centerforastro; Intel @usnavy
Joined August 2014
- Tweets 5,737
- Following 2,592
- Followers 1,259
- Likes 19,242
141 Photos and videos
Elizabeth Hudson retweeted
Jun 10
interesting segment on bio in Dario's essay. specifically the part below. I'd be very curious how he imagines AI can be used for PD/PK modeling and tox.
I assume he means "ask claude for whether a molecule will be toxic" but the existing data on the web for drug tox is probably not sufficient to be good at this, and only reaching human med-chemist-expert level would be effectively random.
would be cool to see them pay to generate a massive new dataset 1000x bigger than anyone else and throw it into the training set though.
10
11
98
19,953
With the right tool, model choice is less important.
A cheaper model gget virus beats an expensive model alone. Reliable science shouldn't depend on having the newest, priciest model. It should depend on solid infrastructure.
We need to build for agents. 🛣️
1
2
9
465
Elizabeth Hudson retweeted
Jun 9
Berkeley Professor Mina Aganagic:
“‘I realized that for students to follow me…I had to start reviewing basic algebra stuff, like fractions.’ The lack of mathematical fluency, Aganagic said, extended even to ‘the meaning of equals in an equation.’”
111
429
3,005
144,686
Elizabeth Hudson retweeted
Very interesting! Another big issue not talked about much here is metadata- a ton of the metadata categories are optional and/or very difficult to fill out correctly. It took me 4 hours to submit samples to SRA! Not to mention the many journals where sequence depo isn’t required
Jun 8
New Science Blog: Why has AI advanced faster in coding than in biology?
To agents, bio databases are like cities built before cars—maddening to drive in because they're designed for different traffic.
How do we build infrastructure agents can use?
anthropic.com/research/agent…
3
10
1,408
Elizabeth Hudson retweeted
Jun 9
Whole concept is really fascinating. Imagine if Google searches had cost money per search and how that would have affected the way we evaluated talent.
Jun 7
it's unclear if tech people yet realize that perspectives on ai usage differ dramatically depending on if you are token rich or token poor.
the idea of looping comes from a place of incredible privilege.
recommending that someone who works at an ai startup with unlimited token budget try looping makes complete sense.
however, that same idea told to someone with only a $2k/mo token budget would seem completely implausible.
what happens when millions of people start evaluating jobs based on what their token allowance will be? what happens to the longtail of administrative paper pushers that don't adapt? what happens to low throughput artists, engineers, or even lawyers and actuaries that can't be trusted to allocate spend?
every form of individual knowledge worker is going to be evaluated from the perspective of "do i trust this person to allocate tokens effectively?"
there are going to be massive investments into new hiring processes. organizations will need to rearchitect their structures to find orchestrators like Peter
my near term recommendation is to do everything in your power to work somewhere that trusts you to spend tokens.
you're ngmi if you only have $2k/mo to play with.
1
1
14
2,374
Elizabeth Hudson retweeted
That goes for about 90% of all biological data. It’s just flat files. Or screenshots in a PDF. We’ve got a long way to go and a major effort to standardize data.
1
3
6
1,130
Elizabeth Hudson retweeted
AI labs wanting to solve biology and learning how our data is organized
ALT Hilary Clinton in an East Harlem kitchen meme
18
75
887
58,072
Jun 9
“To agents, bio databases are like cities built before cars—maddening to drive in because they're designed for different traffic.”
“Treating data as secondary to publications got us here”
I'd argue that these observations are all spot on, but that bio databases are hard for agents to traverse in part because they are *also* hard for bioinformaticists to traverse. Treating data as secondary to supplemental to publications got us here - empowering bioinformaticists with agents will help us progress faster
3
255
Jun 8
Attending a conference is definitely the solution.
Jun 8
"The fear that great wealth will have a corrosive effect on kids, infecting them with what some call affluenza, is common among parents, who worry their children will be entitled, lack motivation, spend extravagantly or fail to contribute meaningfully." wsj.com/lifestyle/careers/r3…
118
Elizabeth Hudson retweeted
Jun 6
One of the issues using animals to test lifespan extension drugs is that, by using short-lived strains, you might just end up correcting what makes them short-lived.
If you test in long-lived (e.g., 950 days for a rat) animals, most lifespan extension benefits aren't supported.
6
13
142
12,871
Jun 7
How much sector value is underwritten by the assumption obesity-driven diseases persist? If the value shrinks bc the market disappears, do people respond by making bigger more risky more interesting bets … or?
Jun 7
JUST IN: Eli Lilly says its experimental obesity drug cut sleep apnea severity by 60.6% in a Phase 3 trial.
1
1
231
Elizabeth Hudson retweeted
What I find most notable about this chart is that at its peak the US nuclear buildout was actually faster than the current Chinese one. If we could do that 50 years ago we should be able to do much better today.
58
194
2,206
106,804
Jun 6
Each one was somebody’s baby
Jun 6
Omaha is now a death zone, as seen from above. Men pinned down along five and a half miles of hell. See more at substack.com/@alexkershaw
79
Elizabeth Hudson retweeted
May we live worthy of their sacrifice.
.@SECWAR just met World War II veterans at the Normandy American Cemetery Ceremony.
Their courage and sacrifice helped save the world—America will forever owe them a debt of gratitude.
101
748
5,083
180,182
Jun 6
Once people realize that we don’t have data to describe biology like we have data to describe language there will be a lot of VERY MOTIVATED people looking to generate those data before it’s too late.
Jun 6
Researchers at leading labs have started smoking.
Their logic: AGI will cure cancer before it kills them.
Are you lighting a cigarette because you're that confident the models will cure cancer in time?
No? not AGI-pilled...
126
Elizabeth Hudson retweeted
Jun 6
Today, we remember the heroes of D-Day - June 6, 1944.
Through extraordinary courage and sacrifice, American troops stormed the beaches of Normandy, defeated tyranny, and helped secure freedom for generations to come.
Their legacy lives on. America remains forever grateful. 🇺🇸
895
3,509
12,548
414,404
Jun 6
"This concept that - searching the semantic web - there's enough variance to really understand what happens when you swallow a pill, it transits the gut, it engages a receptor and makes someone feel better or die less...that's I don't think a capability of these current models....these current models and the current data they're trained on aren't going to all of a sudden elaborate new targets that will cure cancer." YES
"...and those datasets are somewhere in the industry with lab notebooks and clinical data..." NO.
The lab notebooks are with a CRO. The machine / protocol data were tossed. The data describing the input material (unless it was a well known cell line) were probably never known, beyond the basics ("liver, healthy").
This is like a sales org saying "the information is in Salesforce" except no; the most important deals were done by text, and there's no record in there of them at all.
AI can't compress a clinical trial. It can't generate the data we don't have yet. Our CEO Neil Kumar pushed back on the AI hype with @bloomberg at #MIGlobal with @MilkenInstitute – not because the tools aren't valuable, but because the years still take years.
192
Elizabeth Hudson retweeted
Jun 3
this is an interesting point in the new ted chiang piece – no one really claims that alphafold is conscious, or that sora or midjourney or dall-e are conscious
162
434
4,158
323,752
Jun 4
Both the Dwarkesh and the Vance pod - @jacobkimmel you’re a great interview. Nice frameworks with enough detail so you both learn and remember something. No point here other than to recognize game. So good.
55