Vin Sachidananda
@vin_sachi
Followers
1K
Following
7K
Media
69
Statuses
1K
VC @TwoSigmaVC Prev: NLP PhD @Stanford, startups, AI Research @Google, @Apple, @Amazon
New York, NY
Joined March 2023
(Unsolicited) Advice for a young PhD student - “This is not vocational training”. I spent most of my afternoon today at Stanford. It’s the week before autumn quarter and 8 years since I came out to the farm with wide eyes. It was as good a time as any to reflect at a place that.
6
16
235
@Austen Lol if they called it the Gaussian distribution would they expect we need a bunch of people named Gauss to be a representative sample?.
5
6
415
If someone does this in NYC, I’ll volunteer 1 day/week teaching the batch how to build things with LLMs.
YC but in Austin instead of SF. Who's building this?.
10
2
143
@vanillaopinions This sounds insane but as I see friends move back near home and peoples parents get older, completely makes sense.
1
0
86
@cremieuxrecueil @realGeorgeHotz @finkd @ladygaga @AndrewYang Is this John's Hopkins CTY? Did this in middle school and can say it was one of the best summer schools for math.
1
1
73
@BasedBeffJezos Emmett is an Eliezer doomer decel, this is much much worse than Mira being at the helm.
2
1
57
@martin_casado Yeah. It’s wild that folks (including some VCs) who don’t have a technical understanding of AI are trying to determine guidelines for safe use. If you don’t practice you shouldn’t preach.
3
6
46
Bad advice. Live where you want to - you’ll work harder, be happier, and have the best shot at succeeding there.
So many YC founders moving to NYC post batch and, when asked why, mention the social and cultural scene. NGMI. SF is less happening, but who cares? There’s (objectively) nowhere better to build, and you should be focused on your startup—not how late the bars stay open.
1
3
42
Yann is a great scientist but historically was very biased against NLP/language models. In 2018, he came to Stanford to debate with Chris Manning on where Deep Learning architectures would go. In the last 2 minutes, he proceeded to make disparaging statements about language.
Here are 10 names who have gone on record to say AGI will be achieved in the next 3-5 years:. - Ilya Sutskever .- Geoffrey Hinton.- Elon Musk.- Sam Altman.- Demis Hassabis.- Dario Amodei.- Ray Kurzweil .- Mustafa Suleyman.- Jensen Huang.- Robin Li. Heard Yann LeCun say otherwise.
1
3
40
Very talented YC founder ends our pitch this morning by showing me an implementation of a language modeling paper I’d written 6 years ago (ancient history in AI time). Made my day, future looks bright.
2
0
33
Ridiculous that people think the Chinese are only capable of psyops and not innovation in AI when faced with chip constraints . Meanwhile, 1/3 of the AI PhD students at our best universities {Stanford, MIT, Cal, CMU} are extremely talented Chinese nationals. Makes no sense.
deepseek is a ccp state psyop + economic warfare to make american ai unprofitable. they are faking the cost was low to justify setting price low and hoping everyone switches to it damage AI competitiveness in the us. dont take the bait.
3
1
33
@Commanders @Verizon That was clearly a first down which would have led to a field goal for the win. Not sure what tape the booth was reviewing.
22
1
28
Talked to researchers and investors about the model layer following Deepseek’s open source release. Some points of consensus:. [Researchers] The models are really that good at a very small fraction of the price compared to OpenAI. It will be the default model (maybe Gemini if.
2
2
25
Gavin Newsom will either veto or sign SB 1047 by Monday. The bill threatens the startup ecosystems in CA, some of the best research institutions (Stanford, Berkeley, UCLA), open source developers, and nonprofits. Reach out to request he veto the bill:
A decision on SB-1047 is due soon. Governor @GavinNewsom has said he's concerned about its "chilling effect, particularly in the open source community". He's right, and I hope he will veto this. If you agree, please like/retweet this to show your support for VETOing SB-1047!.
1
8
24
@VCBrags If your seed investor has never built or sold anything early stage, you are in for a world of pain.
1
0
24
You really gotta hand it to OpenAI for scooping Google the day before I/O.
0
1
22
@kwharrison13 You can also differentiate against 95% of VCs by not giving terrible advice and being empathetic to founders.
2
0
21
@nmasc_ @steph_palazzolo So what I’m hearing is that @eshear was *checks notes* 9th choice for CEO?.
1
1
20
@josemorgado It was great hearing his commentary as they know each others game so closely, this podcast episode is also a gem
0
1
19
@robertghrist Wild, this is prob one of the best things about academia - being able to assess new technology against the body of past work and immediately pass on that knowledge to students.
0
0
21
This is really sad. As an academic, Indian-American and the son of legal immigrants through this designation unfortunate to see that someone (@deedy especially in VC) recommend committing academic fraud to game immigration. Clearly a grifter with no respect for science or the law.
a certain indian-origin influencer popular in tech circles wrote an ultimate guide to US citizenship and straight up said "commit scientific fraud" lol you can't make this up
1
0
20
It's a better time to study {CS, Math, EE, Stats} than ever before. If most code is written by AI would you not want to be able to read/modify/debug it. This is like telling people not to learn arithmetic because calculators exist.
cs grads: dropout now, there is not a single job in CS anymore. AI will take over, this field won’t exist in a year.
2
4
19
Alec Radford shifted the focus of OpenAI from RL to NLP. Back in 2015, I was working on a startup (Celect) as was he (Indico) in the same bldg in Boston on south st. Definitely one of the nicest people I’ve met and had such a great influence on the field - big loss for OpenAI.
Breaking news!. Alec Radford departs OpenAI!. As one of their star researchers, he was first author on GPT, GPT-2, CLIP, and Whisper papers.
1
1
19
@bznotes Agreed and the same advice applies to VCs - to be most productive they should do the work on the front end before a meeting.
1
2
19
This is such a poor take riddled with logical fallacies. An economist’s reductionist viewpoint who doesn’t fully understand the scope of energy production and blames his lack of understanding on the ‘engineer types’. Solar advocates typically make the case for reduced cost.
Nuclear fission is no longer the energy of the future. But for lots of engineer types, it serves as a sort of test case for whether the U.S. government is willing to strangle technology to cater to populist movements. Building more nuclear won't do much to change our actual.
1
2
18
They say VC is an apprenticeship business and so I was very fortunate to have Villi as a mentor. I’d learned what it takes to be a great investor and an even better person from him. Amazing to watch him take the leap, I’d expect Category be an exceptional fund.
Today I am thrilled to announce @Category_VC, an early stage venture firm conceived to provide unparalleled strategic and technical support to founders building enterprise software startups. I am also excited to announce Category Ventures I, our $160 million debut fund dedicated.
0
0
18
My Waymo aggressively honked when a car cut it off, if they add a speaker that berates other drivers it’ll be ready for the east coast.
1
1
16
@BasedBeffJezos Msft and other investors should file within the week. I’d expect Msft to take the weights and compute back v soon.
0
0
14
The best Indian restaurant in SF is Amber India - Zareen’s in Palo Alto is probably the best in the Bay Area.
5
0
15
A very large fraction of people in the Bay Area believe it is some divine mandate. Go to a school you like and meet people you get along with.
0
0
14
No wonder the polls are always so off, even the 'smart' one is using high school stats with the worst software around.
There are some confounders here, but the inflation thing is actually statistically significant! Each additional $100 of inflation in a state since January 2021 predicts a further 1.6 swing against Harris in our polling average vs. the Biden-Trump margin in 2020.
1
0
14
@snehaark The right approach is cropping screenshots of your paper and swapping them into the ppt for the first poster you ever made.
1
0
12
@WinstonLaw Great job. Sad that such a person is the student president of a prominent law school. There should be stronger ethical requirements for those studying law to ensure they uphold our legal systems.
2
1
13
Uninformed take by an uninformed ‘economist’ who thinks CA is representative of the US or the world at large. Noah did his undergrad in physics at Stanford - clearly has forgotten it all. Niche does not mean 20% of current production. He should stick to politicking and clickbait
I like nuclear, but it's a niche product. Solar is the main thing.
2
0
13
Advice I’ve given startups I work with but is broadly useful. Cost should not be your primary value prop, it’ll be a race to the bottom. It’s more difficult to find blue oceans than red or to deliver a superior experience than status quo - but in the end it will be worth it. The.
2
1
10
Heard from a friend that 100% of Stanford AI PhDs are having a miserable time telling recruiters to stop blowing up their inboxes with $750K+ job offers.
Just heard from a friend that 30% of her class at Harvard Business School is having a tough time getting even one job offer.
0
0
11
Wild take from someone who writes about “business strategy”. You can not like him as an individual but to call someone who is likely the greatest entrepreneur “not very good at anything” is delusional.
1
0
11
@Scott_Wiener Regulating AI does not “promote innovation”, it’ll merely drive folks out of CA.
0
0
11
Information theory is probably the most useful course that is not commonly taken by AI phds and imo should be required in the same way as linear algebra. From it, we get many useful tools including compression/entropy (perplexity) and distributional shifts (domain adaptation).
1
1
10
(Not investment advice) . Bloodbath in the markets, would expect further declines. Lots of biased folks (investors, politickers) diminishing Deepseek’s work. They showed their work and clear from paper they can innovate at a small fraction of cost compared to American cos.
3
2
10
Many fantastic businesses started out of dissertation research/topics. A few examples (prob missing a lot esp in bio). Google - Larry, Sergey (Stanford).Databricks - Matei (Cal) .Intel - Moore (Caltech), Noyce (MIT).VMWare - Mendel, Diane (Cal).Genentech - Boyer (UCSF).
Why are PhD students so obsessed with starting companies based on their dissertation topics? Like, your PhD project isn’t the pinnacle of your potential—it’s literally just the first manageable project your PI tossed your way. “bUt iP I GeNerAtEd dUrInG mY pHd—”. My brother in.
0
1
10
This is unfortunately very true. From what I saw, there were two types of domestic schooling in PhD programs (1) very expensive private and (2) STEM focused public magnets. The best way to counteract research being reserved for the well-off is investing in STEM public schools.
Great thread/paper analyzing the effect of parental wealth on winning a Nobel prize (spoiler: it is incredibly advantageous to come from a family in the highest 5% of income). If we can create more opportunities for everyone to achieve their full potential, and not just those.
1
0
11
I don’t think I could be any more jealous. > stem magnet hs.> very rigorous stem college.> builds most important library of all time.> retreats from society to do sick math.
This is exactly the LinkedIn you'd expect from someone who discovered the world's largest prime number.
0
1
11
@Jannikschg Yeah this is correct, my dad had switched from physics undergrad to EECS grad school in the late 80’s for this reason. Progress has been v limited.
1
0
10
Turns out you don’t need to be in the Bay Area and deal with the bitter condescension . >60% of all unicorns are not there and NYC is close enough (133 vs 190).
If you're in tech and not in the Bay Area, you lack either judgment or ambition.
1
0
11
To get a sense for why language models did end up winning in 2018 architecturally, it came down to their generality and scalability. Michael Bronstein’s (@mmbronstein) geometric deep learning talk is particularly insightful for providing intuition on the generality of.
Yann is a great scientist but historically was very biased against NLP/language models. In 2018, he came to Stanford to debate with Chris Manning on where Deep Learning architectures would go. In the last 2 minutes, he proceeded to make disparaging statements about language.
1
4
11
Was at a conference last week and spoke to someone whose husband worked in space flight. Discussed how SpaceX could be >$10T in the long run by subsuming telco, space and parts of defense. Imo it is the most important private co w no competition in sight - would buy if I could.
spacex would be a trillion dollar company right now if it were public.
3
1
10
It's really not that complicated
That’s not what he said. What he said was “I don’t even know what a venture capitalist does most of the time.” Which is what most people in the startup world would say about VCs too.
0
0
10
@wagieeacc @AnthropicAI Anthropic is the group of people from OpenAI that would have stood behind Ilya.
0
0
10
@fluxtheorist For sure, he’s brilliant and a giant in the field. Just saying that despite studying linguistics in undergrad he prob had as much/more exposure to math/physics than most people that studied those fields in college.
1
0
9
The same people who say we have this type of STEM talent (IMO, IPhO, etc) domestically are trying to defund our public education systems.
0
1
9
The people I’ve seen who sought out ayahuasca (or psychedelics more broadly) to improve mental well-being would have been *much much* better off eating clean, lifting weights, and increasing social interaction. Those things work better than any pharma drug as well.
3
0
8
@Anthony_Bonato NLP: Attention is all you need.Signal processing (recent): Compressed Sensing.Optimization (recent): Gradient Approx methods {Adam, Adagrad}.
0
0
10
Freestyle is building the ‘Vercel’ for backend - highly opinionated and thoughtfully designed. One of the most ambitious companies I’ve seen this year. Thrilled to support Ben and Jacob - check out the project here:
Cloudstate runtime is the most different runtime in the JS Runtime wars. We took the data in JavaScript classes and made it so instead of being stored on the heap, its offloaded to a persistent data store. When you access data on classes, we pull the data back out of the store.
1
2
10
This type of discourse is blatantly racist and self defeating. It’s without question that we should be welcoming as many brilliant computer scientists, physicists, engineers and mathematicians as we can. Doing so has been one of our strongest virtues and ensures that we maintain.
Watching Elon get ratioed badly on his own site over the H1B issue is a fascinating revelation of reality. 1. He doesn’t have his finger on the scale like the previous regime. People strongly disagreeing with him is actually encouraged. 2. MAGA is decidedly NOT in thrall to.
0
0
10
Only two ways to make good decisions - either from first principles or pattern matching. Former is better than the latter, lower sample complexity.
3
1
10
@Commanders 3 last minute game winning drives and an OT win - nerve wracking but great season so far.
1
1
10
Bad advice. US investors typically have a higher risk appetite and most have European startups in their portfolios. Also, they are not all on vacation for 4 months in the summer.
Caught up with a few early stage founders in London today. All planning to head to the US to fundraise. I advised against it. The market is also tough in the US right now and little incentive for local VCs to look internationally. Plus….all the US VCs are vacationing in.
1
0
9
Makes sense. The extremist progressive policies on drugs, crime, and illegal immigration that Democrats have pushed mostly affects lower income inner-city neighborhoods. It’s easy for those in privileged communities to be out of touch with working class voters and lack empathy.
This is a map of the shift to Trump in NYC. The surge didn’t happen in places like the Upper West Side or Park Slope. It happened in Washington Heights, the West Bronx, Flushing, Richmond Hill, Bensonhurst—working class communities of color. 1/2
0
0
8
This virtue signaling is done by those who care least about the less fortunate (SF is the prime example). Merely a way for privileged folks in tech/finance (mostly in the Bay Area) to feign morality while continuing to drive inequality.
I had a PM at Facebook in 2022 who was reported to HR by a new grad and reprimanded. His offense? . He said "my WIFE" instead of "my partner.". This is when I realized wokeness was out of control.
0
0
9
@McaleerStephen So much trial and error, the seed would effect things a lot. Vividly remember plotting gradient norms to set the clipping param for LSTMs.
2
0
8
Recently I've heard from some smart folks that they'll switch from X to other platforms . Clearly not what advertisers are doing following the election. The reason to leave is that you don't want to see views representative of the general American population. Ivory tower syndrome.
🚨MAJOR COMPANIES RETURN TO ADVERTISING ON 𝕏. IBM, Disney, Comcast, Discovery, Warner Bros, and Lionsgate Entertainment have reportedly resumed ad campaigns on 𝕏 after an almost year-long boycott. The return is attributed to @lindayaX leadership, as she rebuilds trust and ad
0
1
9
The greatest living mathematician is optimistic in the potential of LLMs (with tool use) for theorem solving. What a time to be alive.
1
1
8
Thrilled to support the team at Distributional on their Series A . There’s fundamental challenges building LLM testing platforms akin to those use by model providers (ie Meta, OpenAI). Distributional is bringing this platform to all enterprises. More here:
Enterprise AI adoption is at an all-time high, but so are the risks. Distributional's $19M Series A, which we're excited to have led, brings much-needed automated testing to AI systems!. Read about why we think @dbnlAI can bridge the AI confidence gap here:
0
0
9
This is what happens when the DNC:. - hides the mental illness of an incumbent.- puts forth a terrible candidate undemocratically .- becomes overrun by progressives that have wreaked havoc on the west coast.- fails to listen to the needs of working class Americans.
BREAKING: With almost all votes counted, every state has shifted toward the Republican Party.
1
0
9
I’m sure the CS grads you fund must find this insightful. Also, please tell us how this compares to a ‘Social Studies’ degree in the early 21st century?
AI means Your Computer Science Degree. Is Probably Worth The Same as a Typing Degree in the early 20th century…
1
0
9
@emilychangtv Failure of a board. They let their imagination of safety threats get the best of them.
0
0
7
@mickeyxfriedman Yeah, Geoff Hinton, Ilya Sutskevar and Alex Krishevsky were instrumental in getting Deep Learning to work with Nvidia GPUs for AlexNet. They’re among maybe the top ~100 people to accelerate the field.
0
0
9
Recently spoke with sales and AI/Eng candidates for some portfolio companies. These are prob the two best to channel for interviewer vibes
0
0
8
@nabeelqu Agreed. Nvidia and Alphabet were founded out of Stanford by grad students. Apple was founded by Palo Alto natives. Msft was founded by Seattle natives. Amazon and Meta were purposefully started in the west by founders that were living on the east coast.
0
0
5
The least intelligent person you know learned about Jevon’s paradox this weekend to justify their NVDA calls.
0
0
7
Very grateful to have had the opportunity to support the amazing folks at SDF through their acquisition by dbt Labs. SDF has built a best in class data product that will soon be in the hands of thousands of companies - hard to overstate the impact and remarkably thrilled for.
We’re thrilled to announce that SDF is joining forces with @dbt_labs! Together, we’re setting out to transform data development and analytics for the better. Read more about this exciting milestone here:
0
0
9
This has been truly inspirational to see unfold. Earlier this year, Marc Tessier-Lavigne had pitched us on Xaira - leveraging some of the ground breaking research David Baker and his collaborators had pioneered. We’d invested in the team, science and vision. Last Friday, I’d.
Congratulations to our co-founder, David Baker, and to John Jumper and @demishassabis of Google DeepMind, for sharing this year’s Nobel Prize in Chemistry! Their stunning breakthroughs have provided the foundation for our work at Xaira, and we couldn't be more proud.
0
1
9
@thegautamkamath This is great. I remember talking to a friend who had gone to TJHS in Fairfax who had taken ML and distributed systems courses there - feel like it will be heavily represented.
0
0
9
The best of times are when you are surrounded by people much smarter than yourself and realize how little you know. There’s nothing better.
@Muennighoff @Stanford Nice! I'd rewind time for another run, it's probably my happiest overall era, though often in a type 2 fun kind of way. Have fun! :).
0
0
9
Newsom will veto or sign SB 1047 within a week with the 9/30 deadline. Spoke to a portfolio founder, a prominent researcher, this morning. SB 1047 handicaps ecosystems we care deeply about (OSS, academia, startups, non-profits) in CA. Newsom acknowledges this and should veto
0
1
8
@APompliano Looked into this, sadly seems mostly preventable (obesity: -7 years, smoking: -10 years)
1
3
7
Nothing more wholesome than seeing multiple friend groups hanging together on find my.
0
0
8
Not investment advice, NVDA guiding 7% QoQ growth for Q4 does not vibe well with their 50 fwd P/E.
0
1
7
Amazing to see a number of TechBio startups (Xaira, Isomorphic, Chai, Iambic, Evolutionary Scale) make advancements in foundation models for structure prediction, quantum chemistry, assay prediction, and molecular design. Lots of opportunity to substantially drive down R&D.
We’re excited to introduce @ChaiDiscovery and release Chai-1, a foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of drug discovery tasks. We're releasing inference code, weights & a web interface:
0
2
7