
President & CEO @ycombinator —Founder @garryslist—Creator of GStack & GBrain—designer/engineer who helps founders—SF Dem accelerating the boom loop
San Francisco, CA
Joined January 2008
- Tweets76.5K
- Following6.1K
- Followers1.1M
- Likes254K
Pinned Tweet
Tech gave me everything I have
Its capacity to lift people into abundance is incredible and there is nothing like it
We must make that into prosperity for everyone
"I realized tech is this thing that can bring people out of whatever situation they're in and often into prosperity. And that's what I want for everyone."
@ycombinator’s @garrytan tells @emilychangtv how tech changed his family's life.
Watch here: trib.al/sxg1VGR
Garry Tan retweeted
BREAKING: Nvidia, $NVDA, sets a new record for the largest-ever loss of market cap in a stock in 45 days.
The stock has now lost -$2.8 trillion in market cap from its high seen on October 6th.
There have been 33 trading days since October 6th, meaning Nvidia has lost an average of -$85 billion per day, for 33 trading days in a row.
Furthermore, Nvidia has accounted for ~40% of the S&P 500's market cap losses since its record high on October 6th.
That's one way to make history.
Garry Tan retweeted
one of my favorite pods we’ve done in awhile! @dhh is a legend of open source, who’s still building cutting-edge projects like @omarchy to this day, and it’s actually delightful and often hilarious to hear him chat with @martin_casado and @illscience.
some of my takeaways:
- old infra can become newly competitive because the entity operating it evolves. the things that made linux difficult for humans, for example, are exactly what make the os attractive to agents (which is why dhh is building a modern linux distro called omarchy). terminal-heavy interfaces, obscure config files, and detailed error messages are “the three things agents love” in the words of dhh. which is why dhh thinks the days of microsoft and especially apple are numbered.
- dhh can afford essentially whatever token budget he wants, but says most weeks a single $200 claude subscription covers everything he does. i’ve met a lot of devs in the bay area who like to flex how high their token spend is, so this was actually a super surprising detail to me. dhh owns three subscriptions, and remembers only two weeks in the preceding three months when he maxxed those out.
- dhh says he never wanted to exchange a company of roughly 60 people for one with 6,000 or 60,000. the prospect of that much productive capacity through agents, without the committee meetings and coordination, interests him enormously. you can see how someone could become much more ambitious about the product while retaining the same preferences about how they want to spend their day.
- there are two kinds of “ambition inflation”: one is the ever-increasing floor on what constitutes a noteworthy project in the age of superintelligence. people used to stop in their tracks whenever a cool open-source 3d build or handcrafted game was released on x…now the bar to wow people is much higher. there’s also the ability to exactingly specify your precise vision, and the desire to get pixel-perfect results. dhh has both kinds of ambition omarchy's screensaver runs ascii art through forty demo effects, deliberately recalling the 1990s demoscene. dhh added another animation while a local model loads, purely because it makes the computer look cool. he also wants to see omarchy surpass macos in terms of users, a reasonable task, he thinks, because macos has something like “only 120m users.”
check out the full pod in the qt!
Ruby on Rails creator David Heinemeier Hansson says AI is giving small teams a chance to compete with software ecosystems built by thousands of developers. In his view, as building software gets faster and cheaper, taste and craft become more important.
DHH joins a16z General Partners Martin Casado and Anish Acharya to explain why he's building Omarchy, an agent-first Linux distribution. They also cover why Linux is unusually well suited to agents, the importance of open models, the return of individual hackers shaping the future of computing, and more.
00:00 Intro
00:48 From Rails and Basecamp to Linux
03:58 Why AI agents make Linux the better OS
08:17 Can a third mobile platform happen?
16:02 Does taste still matter when AI writes code?
22:57 What counts as superintelligence?
25:58 Big AI labs vs. open source
31:50 DHH's plan to be bigger than macOS
38:24 Who to follow, and the AI culture war
44:35 The return of the pseudonymous hacker
47:33 90s demo culture and computers with character
YouTube: youtu.be/TtOoLanFzGw
@dhh @martin_casado @illscience
Garry Tan retweeted
Backed by YC and tested by NBA star Moses Moody, our glasses tell you when your brain peaks.
Your ring tracks your heart. Your watch tracks your sleep. Nothing tracks the eyes and brain running the show.
STASIS reads your 100,000+ daily eye movements to tell you:
• When you’re at peak focus
• When your eyes start burning out
• When scrolling and blue light starts costing you sleep
6 sensors. 11mm frame. 92% accuracy. No external cameras.
Pre-orders are open. Get yours.
wearstasis.com/checkout
Garry Tan retweeted
Bessent is drifting way out of his expertise and it looks really embarrassing.
The most well known AI attack was committed by a closed model and “Chinese models” were used for defense.
Garry Tan retweeted
here's a list of mistakes made by the nvidia board contributing to the 40% stock decline:
1) they rushed the decision
2) they made the wrong decision - should have put Jensen on leave to complete the investigations
3) when they made their decision, they didn't communicate the rationale well enough
4) their communications have been inconsistent - they called it a voluntary resignation and then doubled down at every opportunity saying that jensen can't come back (how are those 2 things both true?)
5) they chose the wrong interim leader - any ceo for nvidia, temporary or permanent needs to be a) an engineer and b) inspiring enough to be the glue for the company. im a huge fan of colette kress in her cfo role. but based on earnings calls interviews and all hands, she is not the best pick for this moment. not instilling confidence.
6) in launching a ceo search they are looking external. as bezos mentioned, this should be a last resort.
7) they hosted an all hands and didn't take any questions. isn't the point of an all hands to answer questions and inform the team of what's going on?
8) and now, it looks like they want to try to pay people to stay at nvidia, further diluting shareholders. mind you, these are people who are already independently wealthy, and no longer motivated by money. top talent is motivated by winning and having positive impact on the world. this further shows how out of touch they are.
im probably missing some of the missteps but these are the obvious ones above. and they are contributing to the 40% stock decline. nvidia just had the best quarter in the history of business, is growing 75% YoY, and they launched a massive buyback and historic low prices. and yet the stock is still crashing. i think that is telling you what shareholders demand. better decision making. maybe these board members should look inward.
Garry Tan retweeted
Clearer thinking than 99.9% of VCs had in 2023
Chapeau.
In 2023, Aston Kutcher was asked about why he invested Anthropic at a ‘HIGH’ valuation of $4 Billion.
Ashton: "I think these models look lot like web 1 models like Yahoo, Google and none of these AI companies have really even started reporting their revenues yet."
"I think every company will be AI company, like 10 years ago if you were in commerce and didn't invest in E-Commerce, you aren't even a company anymore."
"And if there 2-3 companies at the foundational layer, the TAM is so large that marking value of a company based on anything other than the potential would be wrong."
Today, Anthropic is valued at a whopping 2 Trillion Dollars.🔥
Garry Tan retweeted
WTF… another Millennium Prize Problem might have just fallen, this time to Anthropic.
Levent Alpöge (mathematician at Anthropic) says he and Claude have discovered a proof for the Riemann Hypothesis AND Generalized Riemann Hypothesis.
unsolved since 1859 btw.
You don’t need an asset seizure tax
You need to encourage building new ideas and startups in California
And then you need a government that spends it wisely on things that help the people
That’s it. The ivory tower academics and the SEIU-HNW union bosses have it wrong.
Garry Tan retweeted
True for building all good products
You will notice that Gemini Carbon thinks differently than Claude and GPT. It often comes up with entirely different but more clever solutions than competitor models.
This is precisely because we never distilled other models. Distilling other models creates a poor imitation rather than unique intelligence.
Garry Tan retweeted
Ok I do have a view on NVIDIA without Jensen.
I’ll tell you the real reason $NVDA will likely fail without him and why the frontier labs and hyperscalers are the biggest winners. (Elon has been in a really good mood recently. Have you noticed?)
The biggest threat to NVIDIA is its own customers, the frontier labs. NVIDIA has to become a frontier lab faster than frontier labs are becoming chip companies. And with Jensen out, they likely won't maintain the momentum to get there.
With recursive self-improvement (RSI), every frontier lab and hyperscaler must become a chip company or lose the AI race. It's that simple, and they've all laid the groundwork (Google TPUs, OpenAI and Meta building custom ASICs with Broadcom, Anthropic building an in-house silicon team, and SpaceX/xAI/Tesla building Terafab).
That's because software-only RSI has clear physical limits. Software can reduce how much computation and data movement a model needs, but it can't change the chip's physical bandwidth, memory capacity, or interconnect. Custom silicon gives labs another way to improve performance: designing the memory system, compute, and networking around their own workloads.
To get the real compounding effect of RSI, you have to co-design the model and the silicon so it becomes a two-way feedback loop: smarter models design better custom chips, and those chips train and run the next model faster. You can't do that on off-the-shelf NVIDIA GPUs because you get zero proprietary edge buying the same chip as everyone else, and getting your next-gen architecture into NVIDIA's silicon means sharing your roadmap with them years before tape-out (which they then sell to your rivals). Plus, AI now automates writing compilers and porting CUDA kernels to other chips, which kills the CUDA moat.
Jensen argued custom ASICs aren't more cost-efficient than GPUs because models change too fast for fixed hardware, and partners like Broadcom charge fat margins too. But hardware doesn't fall behind model changes when AI co-designs the next model and chip together instead of humans designing them separately (which Google and OpenAI are already doing). And labs aren't building custom silicon just to save on NVIDIA's 75%+ gross margins. Closing that hardware-software loop is the main goal, though the cost savings are still massive since RSI requires 24/7 internal inference where custom ASICs are way cheaper and more power-efficient than general-purpose GPUs.
That's why NVIDIA has been trying so hard to become a frontier open-weight lab, buying Groq (fast inference IP), Poolside + Nemotron (internal coding models), and Hugging Face (distribution). If closed labs win, enterprises just call their APIs running on the labs' own custom silicon and NVIDIA gets cut out. Subsidizing free open-weight models is the only way to make enterprises buy or rent NVIDIA GPUs to run the models themselves.
With Jensen gone, it's unclear if the Hugging Face deal even survives DOJ approval (can a GPU company own the whole stack?), and a non-founder is unlikely to take the massive risks needed to become a frontier lab: putting billions of capex into internal clusters and free open-weight models instead of just selling GPUs, while competing directly with its own biggest customers $GOOG, $META, $SPCX, $ANT, $MSFT
No view on $NVDA but If you've been buying it during this sell-off but didn't bother to buy in march or much of last year when the stock traded lower, you'll have to explain how the fundamentals are better today than they were then.
Garry Tan retweeted
it took Resend 617 days to reach $1M in ARR.
last week, we added $1M ARR in just 3 days.
no big contract. just product-led growth.
Garry Tan retweeted
Brad Gerstner says there may not be a violation of law in the Jensen Huang case, and a lot of people now have an appetite to go after Nvidia's board
"I agree with Elon and Palmer that it's crazy to take this person off the floor unless he violated national security laws, right? If he did that, let me be absolutely clear: Jensen is not above the law."
"It's increasingly clear to me that there may not be a violation of law. And if there's not a violation of law, at some point, this board is going to have to explain why they did what they did."
"And I will tell you this: based upon the inbounds to me since we wrote the letter, there are a lot of people who will have an appetite to go after this board."
Sam Altman says he doesn't know if what Anthropic is doing is safe, hours before Anthropic prices its IPO
[ Andrew: Do you think that what Anthropic is doing, the way they're doing it, is safe? ]
"This gets to what Greg is saying. I don't know, and I think they also don't really know what we're doing. And the race dynamic in the fog of war, I think, is a bad thing."
"I think this is a time where more caution, not less caution, is warranted."
"And the dynamics where people are like, well, for whatever reason I need to go max speed here, it is a little unsettling, and I think that this could be addressed in a very positive way."
Garry Tan retweeted
There’s a way of thinking about the ethics of building LLMs which deserves much more attention:
During pretraining of LLMs we distill vast amounts of humanity’s collective work from the internet into the model’s weights. When that knowledge was created, we were not aware that LLMs would be possible. The content was released for others to read freely, essentially as a common good. This includes not only all of our writing, but also the vast amounts of open source code on platforms like GitHub.
Once LLMs were discovered, AI companies were able to access the content easily (in the trillions of tokens), train powerful models and then sell them back to us for hundreds of billions of dollars. They took something we released for free for the common good, and then turned it into trillion-dollar companies to primarily benefit themselves. Coding models in particular, trained from GitHub, are very useful and highly profitable.
At the same time, I believe that we can’t make a strong copyright claim on all of the training data. LLMs, similar to humans, do not perfectly reproduce the source text, but rather learn from it and then synthesize new content. If we wanted to forbid LLMs from learning from it, we would need to answer why humans are allowed to do the same.
There is an elegant way to resolve this tension: We should return the resulting models to the commons, releasing them openly for anyone to download and use freely. This way the data is not repurposed for a few people’s profit, but merely transformed into a more useful form for society. This is the core idea behind open-weight models, which are gathering momentum and are starting to rival the models of proprietary AI labs.
After working in the AI industry for many years, building many LLMs, I’ve come to strongly believe that this is the right point of view.
Although most people today don’t seem to think along these lines, I believe that in the fullness of time this will become the majority opinion and the current state of AI development will look odd in retrospect. Provided the models haven’t taken over at that point.
Your AI can keep working after you close your laptop. Here's what customers have told us theirs do overnight:
- A startup's AI files every sales call each evening, then writes up what's live and what's still waiting to ship. 850+ pages in about 3 weeks, and the founders didn't type one.
- An operations manager wakes up to his plans already updated from yesterday's email, with each change waiting for his OK. His weekly meeting deck builds itself the night before.
- A four-person company built a crew of AI analysts. Every week they track sales and draft the founder's weekly review.
- A small business got 6,000+ emails and ~600 recorded meetings into memory in a couple days.
You set yours up in one sentence.
gbrain.io/docs/workspace/wha…
Garry Tan retweeted
The level of orchestration here is just incredible
A user asked Perplexity to find a suitable parcel of land for a 12-cabin build near Sacramento.
Perplexity found the @mireyehq API, spent $2.20 on 31 requests to Mireye's Land Search and Enrich API, and narrowed 327 parcels to 3 finalists by eliminating parcels with >15% slope, flood zones, wetland zones, or protected land.
Then Perplexity retrieved USGS 3DEP 1m elevation and NAIP aerial imagery, modeled each site in three.js, recommended a winning parcel, and created this 4K 60fps video to recap its process.
Garry Tan retweeted
The 90th percentile OpenAI researcher now uses around $23,000 worth of tokens per day, up 105% from September.
We're sharing new data on how AI is changing research at OpenAI.
We find researchers' AI use continues to rise exponentially, and agents are completing longer tasks, but human involvement remains central in research. Researchers are also spending more time on safety and alignment, with human effort in these areas growing faster than in any other individual research area.
This report aims to increase transparency around research acceleration and inform the governance of increasingly capable AI. We believe that AI labs should be required to track and publicly report indicators of recursive self-improvement (RSI): the process by which AI accelerates the development of future AI systems.
Our hope is that this expanded report will help sustain and advance shared measurement norms across the industry.
openai.com/index/research-ac…
Garry Tan retweeted
BREAKING: Over 10,000 Nvidia employees, 24% of its total workforce, have signed a letter to bring back Jensen Huang as CEO.
Bostrom: "if nobody builds it, everyone dies"
Hard agree
one assumption i want to question about AI is that we'll eventually figure out the “best future” and then the remaining task will be getting everyone to agree to it. people want different things, and what they want can change. where more intelligence will help is in making those differences easier to accommodate.
nick bostrom's influence is worth revisiting on anthropic's ipo day. dario, like many of today’s most influential ai founders and researchers, read his work over a decade ago, and bostrom’s book superintelligence helped define questions the industry still faces. i'd like us to engage just as seriously with the possibilities bostrom explores in deep utopia, his subsequent book which asks what we do after we get superintelligence. some notes from my conversation with @eddylazzarin:
-superintelligence could make radically different versions of the “good life” possible. you might want a 200-year lifespan and a village where nobody carries a phone; someone else might want neural implants. someone else might want to live in vice city (either the real version or just a super realistic matrix-sim). ai could help design the economic arrangements that sustain all of this.
-there’s a secondary problem though - even when we have beneficial tech, we may lack the will to diffuse it. waymo and GLP-1s, for example, can save lives (neurosurgeons have compared waymo’s safety results to a treatment so effective that continuing to give patients a placebo would become unethical). and yet evidence of a benefit doesn’t automatically produce the will to act on it.
-the reality described in deep utopia ameliorates this by imagining that our psychology might one day become changeable too. i might want an environment that suits my temperament but i can also go the opposite direction and manipulate myself to become more adventurous or less anxious, in other words, more suited to my environment. the interesting possibility is having more agency over both worlds, external and internal. glps’ effect on appetite already suggests how technology can change the effort involved in pursuing a goal (though again, it is telling that not everyone who *should* be on peptides *are* despite the health benefits). the sci-fi version extends that possibility enormously.
-timing matters. bostrom reminds us that people continue to die while potentially lifesaving advances remain unavailable (he literally says “if nobody builds it, everyone dies” as a hypothetical inversion of yudkowsky/soares’ book).
i’d like ai to make it easier for people to live very different lives, without creating this burden of proof that theirs is the best one.









































