startup investor/helper, founder @conviction. accelerating AI adoption, interested in progress. tech podcast: @nopriorspod
Joined July 2011
- Tweets 14,671
- Following 3,263
- Followers 144,975
- Likes 23,099
1,362 Photos and videos
California’s next set of leaders should take notes from Greg Abbott on how not to chase off a golden goose that can fly to space, create wealth, and build local industrial capacity from scratch
Jun 13
Thanks, Elon.
And thanks to you too, Gavin.
6
6
131
16,090
Jun 12
break out of the prison of the mind. and the earth.
ad astra! congrats SpaceX friends
5
4
140
7,815
Jun 12
The class I loved best in school was architectural history.
I am not religious, but visiting Sagrada Família was one of the most moving experiences of my life.
I am awed by the generations who have built it, and the people and church that have supported it, and by its beauty.
26
4
212
20,076
sarah guo retweeted
Jun 11
We're trying a new experiment at @cursor_ai - interviewing devs we admire.
I chatted with @oneill_c & @part_harry_ from @baseten about how they use coding agents. We discussed their current dev workflows & some predictions for the future.
Check it out below!
31
48
730
106,932
Jun 11
fun recent set of interviews for @NoPriorsPod (Zuck, @satyanadella) and we have @LipBuTan1 (@intel CEO) next week. who do you want to see in the lineup!
18
2
105
23,533
Jun 10
new @NoPriorsPod with Priscilla Chan, Mark Zuckerburg and Alex Rives:
- taking seriously the @biohub mission to cure and prevent all disease (soon!)
- Model release of ESMFold2 and ESM Atlas (beating AlphaFold)
- new biological knowledge from the models
- ecosystem strategy
15
24
372
74,355
Jun 10
Full interview on YouTube or anywhere you get podcasts: youtu.be/u8cYwaLzN-k?si=BgJM…
3
19
7,092
sarah guo retweeted
Jun 10
1/ You can shrink a language model's KV cache by 200×, in a single forward pass, and it still answers correctly.
At 256k context that's 36 GiB of cache down to ~360 MiB, with no change to the base model.
Here's how we did it 👇
27
102
975
109,099
Jun 10
C.S. Lewis is always a good re-read but especially at this time
35
8
282
27,059
sarah guo retweeted
Jun 10
Mark Zuckerberg, in an extensive interview about Biohub and the Chan Zuckerberg Initiative to @saranormous @eladgil:
"The biggest thing that we've formalized is that Biohub is the main focus of our philanthropy. That's been a very big shift."
youtube.com/watch?v=u8cYwaLz…
2
9
6,606
Jun 9
cursor, lovable, cognition numbers all a big narrative violation. wasn’t everything in the path of agi labs (especially the #1 fight, coding agents) supposed to die, not accelerate
60
30
757
125,508
sarah guo retweeted
Today @Meta is proud to launch America’s Workforce Academy with our partners.
This program will provide paid training, certification and a job for Americans of all backgrounds to be part of building American leadership in the world.
Because we believe the Future is for Everyone.
wsj.com/opinion/high-tech-se…
218
389
1,681
1,004,725
Jun 9
1/ thinking: so many default behaviors of models are based on the failures of internet data. secondary framings like rankings, summaries, takes, etc appear far more often in training than the primary sources they compress.
20
4
116
34,018
Jun 9
3/ folks broadly complain about models being “nerfed” but interesting to think about specific bad behaviors in models that I expect might get worse as companies “compute/cost-optimize” their general products
3
23
6,175
Jun 9
4/ unclear if it’s just the upstream pretraining data or the actual reward today of faster time-to-answer/less external call
2
14
3,443
sarah guo retweeted
Jun 8
Introducing FrontierCode: a coding eval that raises the bar for difficulty & quality. Each task took 40 hrs of work by leading open-source maintainers.
Models write sloppy code that works but isn’t maintainable. Our eval is first to measure: would you actually merge this code?
234
313
4,283
2,507,259
Jun 7
just watching the train of thought for any of the frontier class models is simultaneously thrilling and so disappointing. it fails by being generic and it knows it! it’s so repetitive! it cannot juggle enough ideas in its brain at once!
38
4
117
16,399
Jun 7
Scott in 1998: legibility is a central problem in statecraft. states often knew precious little about their subjects, and flattened them so they could count, tax, conscript, and control them
mētis is the counter concept, the value of local expertise
interesting in 2026 too
12
4
82
57,656