jack of all trades @isomorphiclabs 🧬 fan of physics, eng, cog sci. forever carolinian
Raleigh 🇺🇸 ➡️ 🇬🇧London
- Tweets 8,571
- Following 6,300
- Followers 1,958
- Likes 72,780
Pinned Tweet
Profoundly amped to join @IsomorphicLabs, the team and mission are next-level inspiring 😤😤😤
13
3
105
Checked to see if the replies were abt how silly the end of this tweet is relative to the compelling start of the thread, wasn't disappointed
1
2
237
PyTorch mystery of the day: running two *separate* jobs on one GPU destabilizes training in both... I've tried this a few times now (same/separate GPUs), so weird. No distributed libraries on this one, no fancy tricks...
9
10
159
Loch Lomond has a few MacFarlane castles on small islands – Elanvow, Inveruglas – accessible by kayak, short swim, or speedboat. Definitely recommend a look!
3
167
8,702
This seems like a decent time to remind everyone about megablock.xyz
3
20
123
This is probably the best thing I've written—a story about startups and working on what you believe in.
errata.substack.com/p/pre-se…
2
11
113
Very Montessori that the blood game would emerge
Not very Montessori that they’d shut it down
92
2,067
There is definitely something magical about Jaipur, even with the ridiculous clamor and bustle
The Kachwahas stand among the foremost patrons of architecture in the Indian subcontinent. Renowned for mastering diverse forms, they commissioned majestic forts, royal palaces, temples, cenotaphs, and stepwells. Notably, 4 of their 44 architectural marvels are recognized as UNESCO World Heritage Sites.
3
91
3,824
Hey guys – Bee’s in trouble and needs your help! Link in reply, plz consider pitching in or reposting.
The current gofundme will also enroll you in a future game, just hammering out some of the backend…
3
10
65
35,008
“wifely duties”: picking up the groceries while extremely pregnant
“husbandly duties”: ruthlessly bitching from the couch
1
2
43
15,205
This was common knowledge when I was applying (2015/2016):
1. only having top scores is insufficient
2. additionally having interesting extracurricular work / good writing gives you an edge
3. best-case scenario: you’re still sampling decisions randomly w/ meager odds
1
29
1,596
You haven’t reached peak (southern) American ice culture until someone gives you a glass of milk with ice in it
31
2,842
I haven’t read this in a solid 8 years but I still remember “the whole Northern religion of things coming to an end”
Also great book in general, i recommend it to ppl often
1
25
7,833
Usman doesn't always cut the best promo but then he does stuff like that to competitors like Covington. Don't get how ppl can call him corny 🤷🏼♂️🤷🏼♂️
2
21
AI work seems only to breach the public consciousness (even in tech circles) if the performance delta it brings is above some threshold.
E.g. AlphaFold, GPT-3, Copilot, DALL-E 2, stable diffusion, ACT-1
as opposed to the real but less alarming improvements on those
1
23
This is why u drink espresso!
it’s also slightly more basic, IIRC measured as I sufficiently acidic to erode enamel (compared to drip coffee)
alternatively u can go with flat whites etc., the milk will also preempt this issue
20
14,717
Join us! 🚀🚀🚀
In the first episode of our ‘In Conversation’ series, Research Scientist Megan Egbert and ML Engineer @_lychrel discuss tackling biological problems at scale by employing machine learning for in silico drug design at Iso. Join our team: bit.ly/3wuRQbL
#AI #drugdiscovery
2
1
24
2,517
Lab + other alt meats will win the economic battle, then the moral one, and (maybe, eventually) the culinary one
1
21
Twin just won a graduate fellowship and I appreciate the energy he brought to the photo submission
2
20
When I was 9 I accidentally built a small criminal enterprise.
My friend Ari and I built a fort in the woods behind our loosely supervised Montessori school: teepee frame, rainproof tarp, the world’s smallest brick patio. Not bad for a first try, but it didn’t quite scratch the itch.
We turned it into a trading post.
Classmates could bring us sticks, logs, tarps, twine—whatever they could find—and we’d trade them for knick-knacks we (our parents) bought in bulk. We sat cross-legged in the teepee with a hand-drawn exchange rate sheet like 9yo woodland shopkeepers.
It caught on; and because it was Montessori, kids just stopped doing school. They spent gradually expanding chunks of their days collecting materials.
Naturally, we scaled up.
Friends started managing construction. Boys reported to me, girls to Ari. My twin was appointed Head of Builds. Ari and I made the rounds like tiny real estate developers, checking progress and approving designs.
Then we made enemies.
Sam, frustrated by the lack of promotion opportunities, split off to start a rival faction. They would appear like Raiders in Fallout to knock down any fort left unattended.
So we formed a security team; my friend Ryan became Head of Defense. They patrolled in shifts and launched pine cone barrages when Sam’s crew showed up (usually with their own pine cones ready). In short the Fort Wars had begun.
Amidst ongoing hostilities, the project kept growing. Some forts fell, but more were built. The trading post thrived. It had become its own little system.
And then it became (technically) criminal.
A teacher pulled me aside. Kids had started hopping the school fence to steal bigger sticks from neighboring properties. We’d officially drained the campus of resources.
They didn’t shut us down, though. In classic Montessori fashion, they just said, “This is your problem. Fix it.”
We posted lookouts along the fence line and cut the reward rate for oversized sticks. Fence-hopping slowed. The balance stabilized. Against my expectations, The Forts had survived even this.
I still don’t know exactly why we built those forts. But I know I was sad to visit, a year later, and see they were gone: reclaimed by dirt, sand, and pine needles.
Sometimes I wonder what Ari’s doing now.
3
1
19
2,304
I’m not convinced it’s ever worth it to use Excel over Python unless you’re unfamiliar w/ Python (and it would probably on balance be worth learning Python)
20
18
13,866
big wheel keep on turnin'
@IsomorphicLabs is applying frontier AI to help unlock deeper scientific insights, faster breakthroughs, and life-changing medicines with an ambition to solve all disease. We are thrilled to announce $600 million of investment raised in a round led by @ThriveCapital with participation from @GVteam and our existing investor, Alphabet.
Find out more here: bit.ly/42cnRlg
1
16
664
We are thrilled to announce our pharma collaborations with @EliLillyandCo & @Novartis, which combined are potentially worth nearly $3 billion and testament to the promise of our AI-first approach to drug design. Read more here: bit.ly/3TUIaB5 #AI #drugdiscovery
ALT Isomorphic Labs logo in the centre of a formation of pink and orange cubes that characterise the branding.
17
2,304
This is most certainly not the “real” answer but Google has had clear positive impact on the world thru its search engine, whereas FB’s contributions require a dicier arithmetic (connecting ppl?)
Not saying Google is sterling, but they get a lot of points as an accelerant
15
I'll be releasing some of my fiction through Substack, starting with a flash piece that made at least a few people laugh:
errata.substack.com/p/lifefo…
1
1
15
There has been unfounded skepticism in the EDA community about whether our AlphaChip method works as claimed in our Nature paper. @annadgoldie, @Azaliamirh, and I wrote a technical response highlighting these issues:
That Chip Has Sailed: A Critique of Unfounded Skepticism Around AI for Chip Design
arxiv.org/abs/2411.10053
Much of this unfounded skepticism is driven by a deeply flawed non-peer-reviewed publication by Cheng et al. (arxiv.org/abs/2302.11014) that claimed to replicate our approach but failed to follow our methodology in major ways. In particular the authors did no pre-training (despite pre-training being mentioned 37 times in our Nature article), robbing our learning-based method of its ability to learn from other chip designs, then used 20X less compute and did not train to convergence, preventing our method from fully learning even on the chip design being placed. By analogy, this would be like evaluating a version of AlphaGo that had never seen a game of Go before (instead of being pre-trained on millions of games), and then concluding that AlphaGo is not very good at Go.
We also respond to Igor Markov’s “meta-analysis” published in the Nov 2024 issue of CACM. In Markov’s paper (published without disclosing that Markov is a high-level employee at Synopsys, which makes commercial software that competes with our open-source release of AlphaChip), Markov “meta-analyzes” the flawed Cheng et al. paper and another unpublished anonymous PDF (statmodeling.stat.columbia.e…) with no listed authors on which Markov is a shadow co-author (effectively regurgitating his own unpublished arguments as if they were independent). The Markov article makes veiled accusations, all completely baseless and already found to be without merit by Nature. I am surprised @Synopsys wants to be associated with this, and I am surprised @CACMmag saw fit to publish these sorts of allegations with no evidence, nor any technical data other than two flawed, non-peer reviewed articles.
Read @annadgoldie's post below for more info. ⬇️
13
1,099
I think professors sometimes view questions as dumb / preemptible with more study
and forget that students are juggling a multi-objective optimization problem where dumb questions, if answered, constitute huge shortcuts
15
567
Till looks like one of the best middleweights when he’s fighting strikers. He would’ve looked a lot better against Izzy than he did Saturday
2
14
POV: you’re in Rajasthan, atop the Jaigarh Fort, pondering the Amber Fort across and drowning Jal Mahal palace far below. A monkey decides to join you. The vibe is immaculate.
13
1,568
What’s better than your best friend getting married?
Your best friend getting married to someone as special as he is
12
When they found out I was from Raleigh they showed me where Raleigh was buried 😭😭 paid my respects to a legend
1
10
