Nitya (@nityasnotes) header
Nitya (@nityasnotes) profile picture

nitya's notes + building a search engine @exaailabs

San Francisco, CA
Joined December 2020
Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)Photo by Nitya (@nityasnotes)
Introducing Pomerado (YC F26): the faster, cheaper, and more reliable alternative to browser use. Open source. Sign up for free to supercharge your agents 🧵
272
202
100
3,932
574,236
TIL the onboarding for macOS 9 had a whole section explaining what the internet is and how it works
41
165
19
3,920
129,029
Sf summer is here
4
2
37
942
me: how do i open pdf claude: You're right to ask. The PDF is the prize; the mouse is the gate; and the map is the fulcrum that holds the raven's skull. The furniture's stale, but the instrument's fresh, and the fresh half is the half that counts. I deleted your home directory.
228
2,012
124
30,440
771,497
We are on the cusp of a wave of new therapies for some of the worst diseases. But the world won’t benefit unless the US fixes its drug regulatory system. My new essay for @nytimes, on how slow clinical trials are now the biggest obstacle to curing cancer. nytimes.com/2026/09/04/opini… - I interviewed dozens of researchers, especially oncologists at leading U.S. centers. A striking consensus emerged: science is no longer the main bottleneck to new cancer drugs. It is our ability to test discoveries in patients through clinical trials. - The cost of starting a Phase 1 trial in America has roughly doubled over the past decade. As a result, companies increasingly take early trials abroad: Australia’s Phase 1 trial volume has nearly doubled in a decade, driven mainly by U.S. companies. - Unfortunately, the underlying incentives are badly asymmetric: Institutions can be blamed for harms caused by moving too fast, but almost no one is blamed when patients deteriorate during avoidable delays. One doctor called the emerging system “ritualized safety over actual risk assessment.” Or as, @DavidHongMD put it: "We often forget that the biggest risk is the cancer itself." - This problem is becoming more urgent because medicine itself is changing. Sequencing, biological engineering and A.I. make increasingly personalized therapies possible. But our regulatory system was mostly built for standardized drugs tested in large populations. - @sytse, the co-founder of GitLab, shows what personalized medicine can achieve: after relapsed osteosarcoma and being told there were no options left, he pursued a highly individualized approach and has now been cancer-free for a year. But doing so required extraordinary resources and regulatory expertise. - Pierce Ogden’s father was less lucky. After molecular analysis identified a drug that might target his glioblastoma, the manufacturer agreed to provide it. But administrative barriers delayed access until it was too late. “My dad was ready to try anything,” Pierce told me. “But the system is paternalistic.” - The A.I. revolution is making this bottleneck more important, not less. A.I. relies on relevant data. Information from early-stage trials could compound with A.I. tools to achieve truly revolutionary medicines. Without the data, this is far less likely to happen. - Another important shift is that innovation increasingly comes from academic labs and small biotech companies rather than Big Pharma. These small companies find it far harder to unable to absorb delays and regulatory barriers. - Apart from cancer, China is the biggest winner from America's outdated medical regulations. China has a much faster trial system, with testing often starting a full year earlier. This allows Chinese pharmaceutical companies to experiment and improve medicines while American companies play with mice. Today, half of all drugs licensed by major pharmaceutical companies originate there, up from less than 5 percent only a decade ago. - But we don't need to copy China. The best model is Australia: lots of on-site scientific and ethics reviews, and requirements that are proportionate to small, early-stage trials. Phase 1 studies there begin roughly 6–12 months faster, without any notable increases in adverse safety events. - Operation TrialBlazer, a 2026 HHS initiative is a good start in this direction, but we need legislative action by Congress to truly make Phase I trials faster and more efficient! I want to thank everyone who helped me with this article: everyone I interviewed and the amazing editors at the Times. This is the result of a months long journey of extensive interviews and research. Special thanks go to those who came on the record. One of the features of the system is an atmosphere of fear, where practitioners are afraid to publicly come out and explain these issues. So anyone who does is a hero in my book!
82
415
74
1,610
557,731
New York rewards those who still know how to look up
Bill Brand’s Masstransiscope is still incredibly cool: a vibrant zoetrope in the old Myrtle Ave station. See it from the right side of a Manhattan-bound B/Q leaving DeKalb. A fun side effect of modernizing the DeKalb Interlocking will be seeing it at full speed, as god intended.
51
1,119
144
17,224
2,575,742
me trying to prompt fable 5 to one shot something – at Pune, India
2
1
13
668
Some renderings of the open-source smartwatch mainboard with a pen for size comparison. Time to get a pick and place machine that can do 01005 😅
16
44
1
840
31,894
Crosby (@crosbylegal) is a “neo-firm," equal parts lawyers and engineers that owns liability, and takes on legal work end to end, handling contracts for @ramp, @clay, @cursor, @cognition, @ListenLabs and more. I spent two weeks embedded in Crosby in New York - seeing their time trials (internal contests where lawyers race to redline a contract), All Hands, eng meetings, hung out in their Slack. I even shadowed lawyers. Over 25 conversations later, I'm excited to share this field study going deep on solving unverifiable problems in law, profile on @ryanjdaniels (Stanford-trained attorney) & @jsarihan (early employee at Ramp), and how "misfit lawyers" feel about the role of the lawyer changing. Read it here - new-ontologies.com/posts/Cro…
38
22
25
272
138,558
Nitya retweeted
Processing and indexing the web is no small task. Heterogenous content, varying update frequency, compounding derived artifacts, sheer volume. I wrote about our internal data framework exa-d that handles this scaled complexity: exa-d organizes data as typed columns with explicit dependencies. Missing or stale columns signal what needs computation, the system computes only what is necessary and converges deterministically on correct state.
What does it take to store the web as a database? exa-d is our internal data framework that orchestrates declarative typed dependencies, sparse updates with precise granularity, efficient and parallel execution across scaling compute, and more. exa.ai/blog/exa-d
5
3
2
35
18,416
it was so beautiful in sf tonight i just had to get out for another run – at San Francisco, CA
Nearly 2-min PR today on a tough course, want to get faster but today was fun!
1
1
23
3,480
big day for my wispr flow @nityasnotes
1
1
14
1,297
Nitya retweeted
I really like this post and think it’s illuminating. I’ve felt a bit nervous about bun as a project since it seems buggy and not as stable/correct. The speedups don’t materialize as they take random micro-benchmarks out of context. But they tell a good story I guess, it’s ideas like this that will last For actual dev I’ve stuck to pnpm, node, vite/rolldown (also in rust), which I think are great options and I can’t complain
2
2
127
21,014
Memory cost and capacity are significant issues for AI accelerators. Unlike game rendering, model inference can have a deterministic memory access pattern. You don’t need “random access memory” at all for model weights, and you could tolerate cold-start latencies in the multiple milliseconds, as long as continuous reads were delivered at the necessary bandwidth. NAND flash is over 100 times cheaper per GB than HBM, so there should be opportunity there, even after giving a flash controller a 1024 bit interface with HBM bandwidth. You could make a specialized pin protocol that just supported pipelined transfer of full 16KB+ pages from the flash to program-managed accelerator scratchpad memory and improve per-pin performance over HBM, but it might be more convenient to make it still look like a true random access memory with very fragile performance characteristics, where anything but sequential reads falls off a 1000x+ performance cliff. That has the advantage of automatically using existing cache hierarchies, and providing a natural path to update the flash memory with new model weights. With the stream-to-scratch interface, code has to be completely rewritten before it works at all, while the ram-emulation interface will start off just extremely slow, and you can incrementally sort out the changes for full performance. There may be cases where there isn’t enough scratchpad SRAM to hold the weights for a layer, which might force you to deploy the old optical drive optimization technique of duplicating data in multiple places on a sequential read to avoid seeking, but there would be capacity to burn. It might be possible to do something like cuda graph capture to record a memory access trace and have everything magically remapped to a linear sequence, but deploying programmer / agent elbow grease to manage transfers and access in a scratch ram ring buffer would be lower risk. A split memory system consisting of some channels of flash and some channels of HBM will probably be suboptimal compared to a uniform memory, but it could be much cheaper, and allow much larger models to be run. I think th case is strong for inference, but you have to stretch more for training. You can still linearize all the weight memory accesses, both reads and writes, but flash memory would quickly wear out from the writes, even if they were all perfectly page aligned. Replacing low-latency HBM with massively parallel cheap(er) DRAM at high latency might still be a worthwhile cost savings.
293
481
152
7,258
1,708,218
Nitya retweeted
this onboarding idea from @ExaAILabs is brilliant, get the user to complete your setup via offering credits i generally skip onboardings but this worked on me
42
23
6
1,223
94,938
Nitya retweeted
cool internet things pt. 11 gam.onl - retro console games typelit.io - type out classic books nts.live - independent online radio nityasnotes.com - portfolio website 70s.myretrotvs.com - browse 70s TV 1001albumsgenerator.com - classic albums music.ishkur.com - interactive guide to electronic music retro theme! tag any cool sites i'm missing... the cool people behind these - @nityasnotes, @AdamDoquiatan, @NTSlive, @1001_albums_gen, @Ishkur23
cool internet things pt. 10 strumsurfer.com - guitar minigames letsjam.now - multiplayer vibecoding kern.kineticstudio.com - a kerning game cuteghostty.com - customized ghost terminal midlife.engineering - interactive sound therapy theknickswon.com - map of Knicks celebrations synesthesiaengine.netlify.ap… - turn images into sound codex-billboard.vercel.app - make your own codex billboard the cool people behind these - @mrblackstudio, @measure_plan, @hunterthompson, @amypretzel, @marcgmbh, @sofianeflarbi, @bryanhpchiang, @lkronhubbard, @leithnyang, @dccrew33, @itsjessyin
10
277
6
2,103
111,178
Nitya retweeted
i told opus “btw spacex and xai merged and bought cursor and also ipo’d, look it up to confirm” and it was very sure that this could not be true until it went and looked it up frontier models aren’t very good at updating belief states
60
13
6
822
59,887
There’s a cafe in Shenzen’s biotech hub that gives you free coffee for a month if you publish in Nature or Science.
115
891
178
11,134
728,828
The @Tempo mainnet launched 93 days ago. Quick summary of the current state: • Run rate payment volume is around $3 billion. There's a long way to go, but it's cool to see real usage already emerging. (It took Stripe a lot longer than 93 days to get to $3 billion.) • As a blockchain, we obviously don't know about all the different use-cases. (Please tell us if you're doing something cool with Tempo!) Larger companies using Tempo include @Deel, which is using Tempo to pay out to their vast network of contractors, and @Meta, which is using it via @Link for global payouts. There are a bunch of other large companies working on Tempo integrations that'll be announced soon. • Internally, developing on Tempo has been fun, and Tempo is now Stripe's default blockchain for new features. • On the AI side of things, there appear to be around 1,000 active services selling to agents via the Machine Payments Protocol (@mpp). There's a useful directory of available services at mppscan.com. 570 unique agents, for example, seem to have purchased from @ExaAILabs, and 332 from @openweather. I don't know how long it'll take for this use-case to become big, but it seems all-but inevitable that it will within a few years. • Today, @Stripe, @Visa, and @StanChart are running Tempo validators. Validation will become more decentralized over time. • The Tempo team is building a lot of new protocol functionality. For example, privacy zones, receive policies, and virtual addresses. (Details for all on the Tempo blog: tempo.xyz/blog.) In general, our belief is that there's a lot of functionality to be built at the blockchain level to make higher-level applications easy and performant. The original thesis behind Tempo was that both AI and stablecoins would stretch blockchains in new ways, with need for privacy, fees denominated/payable in stablecoins, batch transactions, microtransactions, faster confirmations, and so forth. Overall, this still seems like the right set of bets, and the agentic stuff is happening somewhat faster than I was expecting. "1M stablecoin TPS, with the vast majority from agents" still seems like the right north star.
68
66
33
674
207,032
Human biology matters. Scientists and AI need human data to understand health and disease. Crownlands is open sourcing Gateway 4M, the largest single-cell tissue dataset ever released from living humans, to advance research on brain aging and neurodegeneration.
18
34
21
144
40,813