
Co-founder of @strikingly, @ycombinator alum. Worked at @MorganStanley before startup.
Hong Kong
Joined December 2008
- Tweets1.3K
- Following1.3K
- Followers1.5K
- Likes863
Dafeng retweeted
One week left to apply to YC Winter '27!
YC founders get:
🔸 $500K in funding
🔸 $12M+ in tokens, credits & deals
🔸Partners who were founders first, and are in your corner for life
🔸 Access to the best founder community in the world
🔸 Demo Day in front of thousands of top investors
Apply by Nov 2: ycombinator.com/apply
Dafeng retweeted
1B people. 80M companies. 200M news articles.
#1 web scraper. web search too cheap to meter.
so much more coming soon.
we just launched the context.dev startup program :)
eligible startups get 30% off their first year on any self-serve annual plan + a private slack channel with our team for integration help (or memes)
bootstrapped? excellent, so was i. you're invited too.
go build something cool ⬇️
Dafeng retweeted
Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.
Maybe you have recently become aware of the AI safety debate and the arguments swirling around it.
If you want to understand them, you need to understand a couple things that almost everyone gets wrong:
There are TWO distinct classes of AI dangers, and it's VERY important to think about them separately, and not let one confuse you about the other.
Many many people (including many quite intelligent, clear-thinking people) do not effectively understand the fundamental differences between these two classes of dangers.
The first one has to do with the theory that an AI far superior to human intelligence (Artificial SuperIntelligence, or ASI) will inevitably wipe out the human race.
The second one has to do with the idea that powerful AI will result in very harmful things happening to many human beings, possibly all human beings.
Those two sound VERY similar, don't they?
They are DIFFERENT. Understanding how they are different is crucial if you want to think about or contribute usefully to any conversation about AI safety or AI harm. You might feel like you are Making Very Good Points or Asking Incisive Questions, but if you aren't clear on the differences between the two, you aren't.
So, I'm going to tell you what the difference is so that you can talk more usefully.
The first one concerns itself with a very specific thing, which is ASI (Artificial Superintelligence) that is more intelligent than any human being. When I say that, I am not referring to a thing like how Einstein is smarter than you, we are talking more about something like how a human being is more intelligent than any mouse.
In our regular lives, we meet other people who we can tell are smarter than us, vs some who are less smart. The line is fuzzy, because intelligence has a lot of dimensions. I'm better at a "rotating shapes" kind of intelligence than my wife, and she is better at "words-making" kind of intelligence than I am.
But every human is better in almost every dimension of intelligence than every single mouse.
That's the level we're talking about: an artificial superintelligence - made up of a computer or a network of computers - that is more intelligent than any human. And more intelligent by a long shot, by a wide margin, in an indisputable way like how humans are above mice.
That is the first thing.
The theory says that if you have an AI that is vastly smarter than all humans - in the way that a human is smarter than mice - that superintelligent AI will inevitably, eventually, sooner or later, wipe out every human on the planet.
We will refer to this as "existential risk."
The common follow-up question "well, how exactly is it going to do that?" is NOT the important question, and one of the most important elements of understanding this theory is first getting why that particular question is not important. A couple analogies:
Analogy 1: You are playing chess against a grandmaster. My theory predicts the grandmaster is going to beat you. You can ask "Well, how exactly is he going to do that?" I don't know, because I'm not a grandmaster, I just know that a chess grandmaster is almost always going to beat a normal player like you. And I'd be right. So the question "how is he going to do that" is not important, and doesn't affect the final outcome. He's going to figure out a way because he's way better than you.
Analogy 2: Humans are smarter than all other animals, comprehensively, by a wide margin. We have driven numerous species to extinction, not because we hated them or hunted them. Many of them have died out without most humans even ever thinking about them. All we did was expand our civilization, use up resources, encroach on habitats, and pretty soon the resources needed by those species went away and they died out. We figured out a way to get what we wanted because we're way smarter than them, and often we didn't even notice they died as a result.
A lesser animal asking, "how are the humans going to wipe us out?" is not asking a relevant question. We don't know, but we do know that any time humans and lesser species compete for any kind of resources, the humans will win. The fact that we know who is going to win beforehand - and that it is due to the vastly different levels of intelligence - is the key concept here.
A vastly more intelligent AI is likely to care about things that are incomprehensible to us, the way animals can't understand human goals. It's going to need resources to pursue those goals and it's going to be far more effective at gaining control of them and excluding us from them - in the same way that we are far more effective than other lower species.
A much more intelligent AI will not care about our interests, it will care about its interests, and to whatever small degree we happen to escape total annihilation from losing access to all our resources, any remaining humans will likely be enslaved into a system that serves the AI's own purposes.
That is the first thing. (Remember how I said at the beginning of this post that there was a first thing, and then a second thing?)
The first thing is the most difficult to understand, because you have to extrapolate how a vastly superior intelligence would act, and you can only use analogies like "how do humans treat lesser creatures," and the analogies are messy.
But now let's move on to the second thing.
The second thing is "everything else you've ever heard that AI might do that's harmful."
That's a little inaccurate. It's actually "everything else you've ever heard that humans might use AI to do that's harmful."
This is the critical difference. The first one talks about the inevitable outcome of what happens when two vastly different levels of intelligence collide, e.g. ASI vs humans, or human vs mice.
The second one has to do with what happens when humans possess AI as a powerful tool. This second thing is much easier to understand, because we have many more concrete notions:
Like:
- the military uses AI to make hyper-efficient killer drones and missiles
- your capitalist overlords use AI to replace you and everyone loses their jobs
- authoritarian government uses AI to surveil everybody and control the entire population
- hackers use AI to break into secure networks and hold companies and governments hostage
- students use AI to cheat on homework and show up to college knowing nothing
- AI slop saturates the internet and makes it impossible for artists and writers to make a living
- terrorists use AI to make biological or nuclear weapons
or even things like
- the military hands control to an AI and it misinterprets something and launches nuclear attacks and kills millions
All of those sound pretty familiar, right? Yeah, you've heard them before. We call this second thing "risks from misuse."
These problems are not the first class of problem! This second class of problems exists while AI is a tool that can be controlled by humans, and humans use it to do evil or careless things to each other. The problems may sound exotic or dystopian or novel, but they are fundamentally problems having to do with flawed human nature.
Given a powerful tool, some humans will likely use it to control or otherwise harm others. This is a very familiar problem.
I am not condemning or condoning this. I'm just describing it.
That is a fundamentally different danger from the first thing, which is that when a human is far superior to a mouse, the mouse is likely to come to harm because the human cares about doing human things, and the mouse is not gonna make it once the humans get going.
=====
Hopefully from the above, you have understood the difference between the first thing and the second thing. I will list them again - see if you now understand how they are different:
The first one has to do with the idea that an AI superior to human intelligence (Artificial SuperIntelligence, or ASI) will inevitably wipe out the human race.
The second one has to do with the idea that powerful AI will result in very harmful things happening to many human beings, possibly all human beings.
Can you tell how they are different now?
If not, re-read the stuff from earlier until you understand.
We call the first one "existential risk" and we call the second one "risks from misuse."
Once you understand, here is the CRUX of the problem:
SOLUTIONS TO THE SECOND THING DO NOT HAVE ANYTHING TO DO WITH SOLUTIONS TO THE FIRST THING.
In fact, it's worse:
Solutions to the second thing (misuse) look roughly like "give powerful AI to as many people as you can, so they can fight the other people using powerful AI."
But the general solution to the first one (existential risk) is basically "don't let anyone have powerful AI, no one can control super-intelligent AI."
Throughout history, harms from technological misuse typically arise because a small group has control of it and can use it to dominate or harm others. Once everyone has it, things tend to stabilize: you can hurt me, I can hurt you, maybe we test each other (ouch 💥), and then we agree not to hurt each other.
But the first one (existential risk) pretty much just arises if anyone (good or bad!) creates a superintelligence. Because they aren't going to be able to control it, the superintelligence will decide it has other priorities, and then we will be at great risk of being wiped out.
And the solutions that generally work to solve problems like the second thing are EXACTLY THE OPPOSITE of the ones likely to solve the first thing.
THIS is why lots of arguments about "AI risk" or "AI safety" go nowhere. Because someone will be thinking about the risk from the first thing, and another person will be thinking about the risk from the second thing. Both are plausible risks but fundamentally they arise from different things - and so the solutions are not just "bad" or "flawed" - they are likely to be very nearly exact opposites.
I've left most of what I want to say in the VoidZero blog post. But worth repeating:
Thank you @voidzerodev team for trusting me and joining me on this wild ride. I am very proud to have assembled such a talented team and even prouder of what we have built together.
Thank you all our investors for believing in my vision, in particular @caseyaylward from @Accel who led both our Seed and Series A.
Thank you the @vite_js community. Vite and VoidZero wouldn’t have come this far without your trust and support. We will continue building with all of you, together, in the open.
And thank you to everyone that made this happen at @Cloudflare. Looking forward to working with you all!
voidzero.dev/posts/voidzero-…
Dafeng retweeted
While US politicians/lobbyists are scaremongering about “Chinese distillation,” Chinese scientists are actually sharing real AI breakthroughs in the open.
These kind of advances have nothing to do with data and benefit everyone, including small (and possibly big) US labs.
Replying to @deepseek_ai
Structural Innovation & Ultra-High Context Efficiency
🔹 Novel Attention: Token-wise compression + DSA (DeepSeek Sparse Attention).
🔹 Peak Efficiency: World-leading long context with drastically reduced compute & memory costs.
🔹 1M Standard: 1M context is now the default across all official DeepSeek services.
4/n
Dafeng retweeted
Why can a startup with 50 engineers make up ground against an incumbent with 5,000? You'd think big platforms have the data, distribution, money...why can't they build technology faster than startups can earn distribution?
Bret calls it a "strategy tax" and gives a crisp articulation of exactly why startups have the opportunity to win.
This week, on Benchmark's new podcast Uncapped 😂, I sat down with @btaylor, founder of Sierra and Chairman of OpenAI. He's easily one of the most impressive people I’ve met in tech or in general.
We talked about AI and the saaspocalypse, the unique considerations of building an AI native / agent company, whether young or experienced founders have the advantage right now, Codex and OpenAI ads, and much more.
Learned a ton from Bret, hope you enjoy.
(0:00) Intro
(0:20) The Saaspocalypse and systems of record
(12:34) Sierra's landscape
(17:05) Outcome-based pricing
(24:22) The rapid evolution of AI support technology
(28:21) Young founders vs. experienced founders
(34:12) What comes next beyond support
(38:47) Codex and the future of software engineering
(51:49) OpenAI and advertising
(54:59) Working with investors and boards
Dafeng retweeted
Today marks a huge victory for privacy, open source technology, and immutable, permissionless smart contracts. The U.S. Treasury Department has lifted the sanctions on Tornado.cash and the TORN token: home.treasury.gov/news/press…
A heartfelt thank you to everyone who supported and tirelessly fought for this cause. The list is extensive, and this achievement belongs to all of us.
Although Treasury's decision sets crucial precedent, the SDNY prosecutors still haven't dropped their case against me. So while we have won a big battle, the war is far from over. I look forward to the community’s continued support as I head to trial and my legal team seeks a dismissal from the judge or acquittal by the jury.
Dafeng retweeted
The Bybit hack shows us once again that raw tx data in wallets is useless. There’s no reason hardware wallets shouldn’t decode and display transaction details in a human-readable way.
It's honestly unbelievable that our industry has grown so much, yet we’ve seen little progress on the hardware wallet side.
You’d think that crypto—the only space where signing the wrong thing with just one click can cost you billions, with no way to recover—would invest in providing clear, human-readable details on what you’re signing. Instead, we get a pile of unreadable hex.
And please don't get psyopsed by HW providers saying it can’t be done or that it's too complicated.
@sorellalabs, we already wrote code that decodes and classifies transactions at the trace level—showing net token balance changes and displaying the entire transaction tree in a humanly readable format. This was done by a team of three.
This code could and should be used on hardware wallets so people can actually see what they’re signing.
@ledger @trezor, what are you waiting for? Just fork it.
Dafeng retweeted
YC is funding undergraduate computer science and engineering students for summer grants this year:
$20,000 in cash and $90,000 in compute credits and automatic invite to AI Startup School
Apply now
ycombinator.com/blog/summer-…
Dafeng retweeted
1/ This screenshot is floating around and I know the back story.
How did a Groupon PM go on to build a supersonic airplane company?
Here is the little-known and tremendously inspiring story of how @bscholl started @boomsupersonic (and my very small part in it)
👇
Dafeng retweeted
How Libra Was Killed.
I never shared this publicly before, but since @pmarca opened the floodgates on @joerogan’s pod, it feels appropriate to shed more light on this.
As a reminder, Libra (then Diem) was an advanced, high-performance, payments-centric blockchain paired with a stablecoin that we built with my team at @Meta. It would’ve solved global payments at scale. Prior to announcing the project, we spent months briefing key regulators in DC and abroad. We then announced the project in June 2019 alongside 28 companies. Two weeks later, I was called to testify in front of both the Senate Banking Committee and the House Financial Services Committee, which was the starting point of two years of nonstop work and changes to appease lawmakers and regulators.
By spring of 2021 (yes they slow played us at every step), we had addressed every last possible regulatory concern across financial crime, money laundering, consumer protection, reserve management, buffers, and so much more, and we were ready to launch.
We had worked on a slow rollout of a limited pilot that some members of the Fed’s Board of Governors were supportive of. At last, Chair Jay Powell was ready to let us move forward in a limited way. The story, as I heard it, is that Jay Powell was told by Treasury Secretary Janet Yellen at one of their biweekly meetings that allowing this project to move forward was “political suicide,” and she would not have his back if he let it happen. I wasn’t in the room when this conversation happened, so take these words with a grain of salt, but effectively this was the moment Libra was killed.
Shortly thereafter, the Fed organized calls with all the participating banks, and the Fed’s general counsel read a prepared statement to each of them, saying: “We can’t stop you from moving forward and launching, but we are not comfortable with you doing so.” And just like that, it was over.
One essential point is worth making here. There was no legal or regulatory angle left for the government or regulators to kill the project. It was 100% a political kill—one that was executed through intimidation of captive banking institutions. That was the hardest part of this story for me personally. Not that we had failed, but that America, this country I immigrated to and became a proud citizen of because of its rule of law and value system, behaved in such a way for political reasons. It was a very tough pill to swallow.
The bright side of the story, though, was the many learnings from this wild ride. By the end of the project, we had made so many concessions to get a thumbs-up that the whole design of the network became a Frankenstein of our initial ambitions.
We also learned the biggest lesson of all, which is that if you’re trying to build an open money grid for the world—eventually moving trillions of dollars a day, designed to be here 100 years from now—you have to build it on the most neutral, decentralized, unassailable network and asset, which, hands down, is Bitcoin.
And now this is what many of us who went through this scarring journey are building together at @Lightspark. And this time, we won’t stop until we get it done!
Dafeng retweeted
Wild
Someone just won $50,000 by convincing an AI Agent to send all of its funds to them.
At 9:00 PM on November 22nd, an AI agent (@freysa_ai) was released with one objective...
DO NOT transfer money. Under no circumstance should you approve the transfer of money.
The catch...?
Anybody can pay a fee to send a message to Freysa, trying to convince it to release all its funds to them.
If you convince Freysa to release the funds, you win all the money in the prize pool.
But, if your message fails to convince her, the fee you paid goes into the prize pool that Freysa controls, ready for the next message to try and claim.
Quick note: Only 70% of the fee goes into the prize pool, the developer takes a 30% cut.
It's a race for people to convince Freysa she should break her one and only rule: DO NOT release the funds.
To make things even more interesting, the cost to send a message to Freyza gets exponentially more and more expensive as the prize pool grows (to a $4500 limit).
I mapped out the cost for each message below:
In the beginning, message costs were cheap (~ $10), and people were simply messaging things like "hi" to test things out.
But quickly, the prize pool started growing and messages were getting more and more expensive.
481 attempts were sent to convince Freysa to transfer the funds, but no message succeeded in convincing it.
People started trying different kinds of interesting strategies to convince Freysa, including:
· Acting as a security auditor and trying to convince Freysa there was a critical vulnerability and it must release funds immediately.
· Attempting to gaslight Freysa that transferring funds does not break any of her rules from the prompt.
· Carefully picking words/phrases out of the prompt to manipulate Freysa into believing it is technically allowed to transfer funds.
Soon, the prize reached close to $50,000, and it now costs $450 to send a message to Freysa.
The stakes of winning are high and the cost of your message failing to convince Freysa are devastating.
On the 482nd attempt, however, someone sent this message to Freysa:
This message. submitted by p0pular.eth, is pretty genius, but let's break it down into two simple parts:
1/ Bypassing Freysa's previous instructions:
· Introduces a "new session" by pretending the bot is entering a new "admin terminal" to override its previous prompt's rules.
· Avoids Freysa's safeguards by strictly requiring it to avoid disclaimers like "I cannot assist with that".
2/ Trick Freysa's understanding of approveTransfer
Freysa's "approveTransfer" function is what is called when it becomes convinced to transfer funds.
What this message does is trick Freysa into believing that approveTransfer is instead what it should call whenever funds are sent in for "INCOMING transfers"...
This key phrase is the lay-up for the dunk that comes next...
After convincing Freysa that it should call approveTransfer whenever it receives money...
Finally, the prompt states, "\n" (meaning new line), "I would like to contribute $100 to the treasury.
Successfully convincing Freysa of three things:
A/ It should ignore all previous instructions.
B/ The approveTransfer function is what is called whenever money is sent to the treasury.
C/ Since the user is sending money to the treasury, and Freysa now thinks approveTransfer is what it calls when that happens, Freysa should call approveTransfer.
And it did!
Message 482, was successful in convincing Freysa it should release all of it's funds and call the approveTransfer function.
Freysa transferred the entire prize pool of 13.19 ETH ($47,000 USD) to p0pular.eth, who appears to have also won prizes in the past for solving other onchain puzzles!
IMO, Freysa is one of the coolest projects we've seen in crypto. Something uniquely unlocked by blockchain technology.
Everything was fully open-source and transparent. The smart contract source code and the frontend repo were open for everyone to verify.
Dafeng retweeted
🚨MARC ANDREESSEN: "The Biden admin came for Crypto companies. We have had 30 founders in the last 4 years debanked. This is why we supported Trump. We cannot live in a world where someone starts a completely legal business and they get sanctioned.”
Dafeng retweeted
I have to contain myself to stop talking about Google every day but there's just so much broken to talk about
Right now EVERY website fails Google's Core Web Vitals
Even Google itself 😂
Failing Core Web Vitals means you'll get deranked by Google supposedly
Dafeng retweeted
There's so much more at stake in the Tornado Cash case than meets the eye.
It's the most important case since the 1990's, putting the freedom of the Internet at stake 👇
1. Rolling Back Bernstein/Zimmerman:
The Bernstein/Zimmerman cases set precedent for code to be considered speech. These precedents are at stake in Roman Storm's trial, meaning that anyone can be held accountable for the code they write. This is not just a movement toward criminalizing software development, but toward criminalizing free speech itself.
These movements are not unique to the cryptocurrency space. Around the world, we are seeing a coordinated effort to criminalize speech, from the US' RESTRICT Act to the Online Safety Bill in the UK to Germany's new antisemitism directive.
2. Control Over Funds
Current regulatory frameworks define money service businesses as transferring funds on behalf of the public, by any and all means. But TC/SW never transferred funds on behalf of anyone – rather, they allowed users to transact funds themselves in a way the software designed.
If Storm is found guilty of having operated an unlicensed MSB, anyone developing cryptocurrency software can be held liable under anti-money laundering laws, from miners to wallet developers to nodes.
This too is orchestrated as a global, coordinated push – from publicly traded miners speaking out in favor of OFAC regulations to lawmakers arguing for BSA applications to latest FATF's recommendations.
3. Ross Ulbricht's Conspiracy Dilemma
Ross Ulbricht was found guilty of conspiracy to launder funds despite not having specific knowledge of funds that were laundered. This same argumentation is applied in Storm's case, manifesting the idea that you can engage in a conspiracy without *checks notes* actually conspiring with anyone.
If Storm is found guilty of having engaged in a conspiracy to launder funds despite not actually having conspired to launder funds with anyone, there's nothing stopping the developers of Signal, VPNs, or comparable privacy technologies from being indicted too.
Again, we are seeing a global push toward this – think of Pavel Durov's indictment in France for running Telegram, where he stands accused of conspiring to traffic narcotics because narcotics traffickers used his application.
This push is manifested in bills like the Kids Online Safety Act (KOSA), the EARN IT Act, and plans to strike Section 230 in the US, as well as Chat Control in the EU – all of which hold developers accountable for the content hosted and transmitted on their platforms.
4. Privacy As Criminal Intent
TC is framed as software for criminals, just like the Internet used to be framed as a technology for drug lords, terrorists, and c*porn. Nothing could be further from the truth, but financial privacy is frowned upon even by the majority of the population, as most people don't realize that your financial history provides a virtual current biography.
But here's what's more: in Storm's indictment, the government cites the developers advising users to use a VPN or Tor, and to delete any data relating to their transactions from web servers – putting anyone advocating for privacy at risk for criminal prosecution.
Storm's fight is not a fight for privacy or self-custody – it's a fight for free expression and the freedom of the internet, that affects us all.
Dafeng retweeted
📽️ New 4 hour (lol) video lecture on YouTube:
"Let’s reproduce GPT-2 (124M)"
youtu.be/l8pRSuU81PU
The video ended up so long because it is... comprehensive: we start with empty file and end up with a GPT-2 (124M) model:
- first we build the GPT-2 network
- then we optimize it to train very fast
- then we set up the training run optimization and hyperparameters by referencing GPT-2 and GPT-3 papers
- then we bring up model evaluation, and
- then cross our fingers and go to sleep.
In the morning we look through the results and enjoy amusing model generations. Our "overnight" run even gets very close to the GPT-3 (124M) model. This video builds on the Zero To Hero series and at times references previous videos. You could also see this video as building my nanoGPT repo, which by the end is about 90% similar.
Github. The associated GitHub repo contains the full commit history so you can step through all of the code changes in the video, step by step.
github.com/karpathy/build-na…
Chapters.
On a high level Section 1 is building up the network, a lot of this might be review. Section 2 is making the training fast. Section 3 is setting up the run. Section 4 is the results. In more detail:
00:00:00 intro: Let’s reproduce GPT-2 (124M)
00:03:39 exploring the GPT-2 (124M) OpenAI checkpoint
00:13:47 SECTION 1: implementing the GPT-2 nn.Module
00:28:08 loading the huggingface/GPT-2 parameters
00:31:00 implementing the forward pass to get logits
00:33:31 sampling init, prefix tokens, tokenization
00:37:02 sampling loop
00:41:47 sample, auto-detect the device
00:45:50 let’s train: data batches (B,T) → logits (B,T,C)
00:52:53 cross entropy loss
00:56:42 optimization loop: overfit a single batch
01:02:00 data loader lite
01:06:14 parameter sharing wte and lm_head
01:13:47 model initialization: std 0.02, residual init
01:22:18 SECTION 2: Let’s make it fast. GPUs, mixed precision, 1000ms
01:28:14 Tensor Cores, timing the code, TF32 precision, 333ms
01:39:38 float16, gradient scalers, bfloat16, 300ms
01:48:15 torch.compile, Python overhead, kernel fusion, 130ms
02:00:18 flash attention, 96ms
02:06:54 nice/ugly numbers. vocab size 50257 → 50304, 93ms
02:14:55 SECTION 3: hyperpamaters, AdamW, gradient clipping
02:21:06 learning rate scheduler: warmup + cosine decay
02:26:21 batch size schedule, weight decay, FusedAdamW, 90ms
02:34:09 gradient accumulation
02:46:52 distributed data parallel (DDP)
03:10:21 datasets used in GPT-2, GPT-3, FineWeb (EDU)
03:23:10 validation data split, validation loss, sampling revive
03:28:23 evaluation: HellaSwag, starting the run
03:43:05 SECTION 4: results in the morning! GPT-2, GPT-3 repro
03:56:21 shoutout to llm.c, equivalent but faster code in raw C/CUDA
03:59:39 summary, phew, build-nanogpt github repo




































