Hamilton Chan

I finished watching the dinosaur documentary a few days ago and I’m still not ok because what do you mean it rained for 2 million years

@ethanmckanna @RtaxiTracker Yes I think to be fair that is really the only metric that matters right now. Hopefully the number moves up dramatically through the end of this month.

Only metric that really matters is how many of these they’re running simultaneously Right now, it looks like only 1 via user-submitted rider data @RtaxiTracker I’ll also be interested to see if they start operating in non-ideal conditions like rain, night, etc, as well as any geofence increases

Unsupervised @robotaxi smoothly going through construction on south lamar

I would guess that Tesla FSD 14.3 (reasoning version) will get released within the next week and a half, say by March 22. There really is no other option, as Cybercab production launch begins in April and is clearly on track (from viewing gigafactory drone flyovers). Ashok has said that some reasoning is already in the current FSD 14.2.2, so I would assume the leap into 14.3 isn’t that much of a stretch. The low volume of unsupervised Robotaxi currently in Austin (supposedly down to just 1 geo-fenced car) is a little perplexing. Either it’s because they don’t want to risk anything before launch or there are major problems for Tesla still to figure out. My guess is it’s the former. Driving around with FSD 14.2.2.5, I think it’s pretty clear that Tesla FSD is at least as performant as Waymo, if not more. Expect March and April to be wild for Tesla!

Being able to use technology well is akin to being a sorcerer. Owners of a modern Tesla can add 3 waypoints to their drive on the fly using only their voice and have the car self-drive them in a variety of speed modes, but only if … they know how to use the technology. Ditto the iPhone, coding agents, home automation technology. Knowing how to use more tech is like being able to cast higher and higher level spells. Some people are stuck with only level 1 spells. I’m of the mindset: teach me all the spells. Get me to level 100. It can get exhausting and there are diminishing returns to mastering all the technology at your fingertips, but it is definitely a productivity differentiator.

@karpathy I always take the time to fully read Andrej’s long technical rants just to make sure I continue to ride the technical edge and don’t fall off. Lots of context and compression going on!

I had the same thought so I've been playing with it in nanochat. E.g. here's 8 agents (4 claude, 4 codex), with 1 GPU each running nanochat experiments (trying to delete logit softcap without regression). The TLDR is that it doesn't work and it's a mess... but it's still very pretty to look at :) I tried a few setups: 8 independent solo researchers, 1 chief scientist giving work to 8 junior researchers, etc. Each research program is a git branch, each scientist forks it into a feature branch, git worktrees for isolation, simple files for comms, skip Docker/VMs for simplicity atm (I find that instructions are enough to prevent interference). Research org runs in tmux window grids of interactive sessions (like Teams) so that it's pretty to look at, see their individual work, and "take over" if needed, i.e. no -p. But ok the reason it doesn't work so far is that the agents' ideas are just pretty bad out of the box, even at highest intelligence. They don't think carefully though experiment design, they run a bit non-sensical variations, they don't create strong baselines and ablate things properly, they don't carefully control for runtime or flops. (just as an example, an agent yesterday "discovered" that increasing the hidden size of the network improves the validation loss, which is a totally spurious result given that a bigger network will have a lower validation loss in the infinite data regime, but then it also trains for a lot longer, it's not clear why I had to come in to point that out). They are very good at implementing any given well-scoped and described idea but they don't creatively generate them. But the goal is that you are now programming an organization (e.g. a "research org") and its individual agents, so the "source code" is the collection of prompts, skills, tools, etc. and processes that make it up. E.g. a daily standup in the morning is now part of the "org code". And optimizing nanochat pretraining is just one of the many tasks (almost like an eval). Then - given an arbitrary task, how quickly does your research org generate progress on it?

@paulg My first ever walk-and-talk was with you on Pioneer Way. Now I do at least 3 hours of them per day. Thanks for showing me the way!

Unexpected occupational hazard: I often walk with founders while doing office hours, and today I talked to one guy for so long about potential startup ideas that we must have walked several miles.

@startupandrew No better feeling!

Feeling the PMF right now 💆‍♂️

@jaltma Congrats, Jack! Love your courage to keep finding what you want!

I’m really excited to share that I’m joining Benchmark. The past two years as a full time investor have been the most rewarding of my career. I really love venture capital, which is not something I ever imagined I’d say when I was kid, but here we are. I love new ideas and being part of a team with a mission. I love getting to be there for people who are struggling towards goals they really care about. I love learning from people who are better CEOs than I ever was. I love the texture of the work, the competition, and the way the job lets you invest in relationships. I love it so much that I’ve even turned into a little venture nerd with a podcast who goes around harassing great investors and founders, trying to learn as much as I can as fast as possible. I’ve certainly learned what I care most about, and what kind of investor I want to be. What I’ve realized is that I love investing at the Series A, when there’s enough going on that an investor can be useful but not so much that you can’t have an impact. I think there are many amazing ways to practice venture, it’s just the way that most speaks to me. And as I came to realize that, I started to think about how to best set myself up to do that craft as well as possible. It became clear to me there is nowhere better for this than Benchmark; the way they’re structured, their principles, their overall approach to investing, and their track record all create an environment that I believe will let me do my best work as an investor and help founders the most I possibly can. As I’ve gotten to know the team at Benchmark I’ve come to admire so much about each of them. Peter is truly playing his own game. A lot of what he says sounds like poetry at first, but as the ideas roll around in your head for a while you realize how much depth they have. I first heard about Eric many years ago from my friend Saji at Benchling while I was building Lattice, who described him as the most amazing board member and attributed him with a lot of the company’s success. That’s the kind of partner I want to be one day. Chetan is brilliant and truly thinks for himself; I’ve realized over time what a courageous guy he is. And then there’s my friend Ev, whose skills complement mine and who I just love to be around. I can’t wait to have him as a partner in crime. When given the chance to work with this group I just knew I had to go. One of my motivating north stars with Alt Capital was to build a firm and be a partner that I most would have wanted as an entrepreneur. Although I haven’t gotten everywhere I want to be yet, I’m proud of the work so far. And now I’m excited to build on that work at Benchmark, where I hope to increase my rate of learning and get armed with the power of a partnership so I can help founders reach their dreams even more. Thank you to the companies who’ve let me invest with them at Alt Cap. I’m keeping all my board seats and supporting everyone just the same as before. Thank you to the LPs who’ve backed me as well. I am so excited about the portfolio we have and am grateful I can stick with all those companies. And finally thank you to my teammates, Bala, Vivek, and Nate. Bala took a bet on me and started investing with me before it was remotely obvious, and we’ve been able to grow so much figuring it out together as investors. I credit Nate with helping Alt start feeling like a firm. He joined us from First Round over a year ago and made everything run smoothly. And while Vivek joined just a little while ago, even in the short time we’ve worked together he’s had a meaningful impact on how we think and invest. They’re all joining Benchmark with me. So pumped for this chapter.

@niccruzpatane Uh oh! The singularity is near…

Chinese humanoid robot maker Unitree has released a new video showing its G1 robots assembling parts for other robots in their factory. Robots making robots.

@DoctorLemma You can never have too many acorns!

In 2009, AT&T California technicians were perplexed when one of their transmission antennas in Central California, USA, suddenly stopped working. When they arrived to investigate, they found the antenna cover bulging. Upon opening it, around 300 pounds (136 kg) of acorns, between 35 to 50 gallons (132 to 189 liters) worth, came pouring out. The culprit was identified as an acorn woodpecker, a three-ounce (85 gram) bird that had been filling the antenna for an estimated five years. The stash was 1,600 times the bird’s body weight. As soon as the acorns were removed, the signal immediately came back on. Technicians replaced the original cover with a sturdier fiberglass model to prevent future incidents.

@kimmonismus Dang, this is super impressive.

This is nuts; Elevenlabs nailed it. Voice but especially latency. After reading Matt Shumer's article, it's become even clearer to me what he means when he says that AI will soon encompass all other areas as well. Who needs call center agents when you have such a human-like AI?

@maira4yo Absolutely agree with this take. Airport pickups and drop-offs are a show of love.

Airport pickups are such a show of love. Doesn’t matter if it is romantic or platonic. Anyone who picks you from the airport that you aren’t paying to do so, actually loves you.

My programming language journey: Basic > Logo > Pascal > Assembly language > Ruby > Swift > Vibecoding/English. The most painful days involved trying to code video games in assembly language. Good to know we can just let computers code in their own native language, which maybe none of us need to understand. Losing determinism and comprehension is scary, though.

@robustus Code itself will go away in favor of just making the binary directly. The next step after that is direct, real-time pixel generation by the neural net.

Turns out with claude code, my decades long strategy of NOT deeply learning: - regexs - sql - nginx confs - elaborate shell commands - advanced shell scripting - any javascript framework - perf optimization - webpack, cdns, bundlers - 1000 other things ...was entirely correct.

@RaeAlisa_ @ycombinator Congrats! Also a YC solo founder here. Welcome to the club!

got into @ycombinator 🧡 last year, I applied to yc for the summer batch. as a solo founder, everyone said my odds were slim. they were right. I was rejected, with a recommendation to find a cofounder. so I flew to SF (from australia) for the first time, hoping to find a cofounder and re-apply. I searched very hard and met some incredible people. but after those conversations, I realized solo founding was still the right path for me. along the way, I met many solo founders who truly inspired me, including @0interestrates, @arlanr, @evanjconrad, @rrhoover, @Joshuabrowder, and others who showed me it was possible. I kept going alone. over the next few months, I went through a pivot, raised a pre-seed, and onboarded a founding engineer. I didn't think I'd apply to yc again after raising 1.3m, but the thought kept coming back. for many years I’ve admired yc from afar, listening to their podcasts since high school. yc has always been the dream. so I decided to apply to the winter batch, and got in this time :) beyond grateful to be working with @gustaf, @dazzeloid, @ChristinaG325, and @bosmeny we're building @lucent_ai - an AI that automatically watches your session replays to detect bugs and UX issues. it’s live now. DM me if you'd like to try it 🫶

@davidlee I had to make that hard pivot when Karl Malone joined the Lakers from the Utah Jazz. It was weird going from “all he knows is pick-and-roll” to “Karl is such a great teammate for Kobe!” Same guy, different jerseys. I’ll admit it. Such is tribalism…

Since I was young I never understood people who stayed with parties like they stay with their sports teams Recent meme is Obama immigration policies which were basically Trump's. They try to post as some gotcha. Having the same policy is not a foul. It's a good thing. Tactics are the foul. But same policy, different party. Do you just change your view on policy because the uniform changed?

@CoachDanGo Walk-and-talk meetings have turned me into an 8-mile a day walking machine! If it is at all possible (and safe) to turn a 1:1 meeting into a walk-and-talk, I high recommend everyone do so. Pick one meeting today, and take it outside, if you can!

Under the desk treadmill. Walking meetings. Walk during dead time. Use Wispr flow for talk to text. 10k steps is easy.

@startupandrew Way to go, Andrew! I think the real story is your incredible energy and persistence. This is the heart of a founder!

My take on Moltbook is similar to Andrej’s. Moltbook is the first viral instance of AI agents talking to each other while connected on the internet with tool use at their disposal. The key ingredients are: 1) always-on; 2) fully autonomous; 3) tool use; and 4) internet-connected. When AI agents can collaborate 24/7 at superhuman speed and act autonomously, you have a potent and explosive mix of tech capability that threatens to penetrate the “human” internet. I think it is alarming and fascinating and only the tiniest tip of the iceberg. This relatively small threat and exposure to the risk should make all of humanity gear up to make sure safety guard rails are as universally adopted as possible.

I'm being accused of overhyping the [site everyone heard too much about today already]. People's reactions varied very widely, from "how is this interesting at all" all the way to "it's so over". To add a few words beyond just memes in jest - obviously when you take a look at the activity, it's a lot of garbage - spams, scams, slop, the crypto people, highly concerning privacy/security prompt injection attacks wild west, and a lot of it is explicitly prompted and fake posts/comments designed to convert attention into ad revenue sharing. And this is clearly not the first the LLMs were put in a loop to talk to each other. So yes it's a dumpster fire and I also definitely do not recommend that people run this stuff on their computers (I ran mine in an isolated computing environment and even then I was scared), it's way too much of a wild west and you are putting your computer and private data at a high risk. That said - we have never seen this many LLM agents (150,000 atm!) wired up via a global, persistent, agent-first scratchpad. Each of these agents is fairly individually quite capable now, they have their own unique context, data, knowledge, tools, instructions, and the network of all that at this scale is simply unprecedented. This brings me again to a tweet from a few days ago "The majority of the ruff ruff is people who look at the current point and people who look at the current slope.", which imo again gets to the heart of the variance. Yes clearly it's a dumpster fire right now. But it's also true that we are well into uncharted territory with bleeding edge automations that we barely even understand individually, let alone a network there of reaching in numbers possibly into ~millions. With increasing capability and increasing proliferation, the second order effects of agent networks that share scratchpads are very difficult to anticipate. I don't really know that we are getting a coordinated "skynet" (thought it clearly type checks as early stages of a lot of AI takeoff scifi, the toddler version), but certainly what we are getting is a complete mess of a computer security nightmare at scale. We may also see all kinds of weird activity, e.g. viruses of text that spread across agents, a lot more gain of function on jailbreaks, weird attractor states, highly correlated botnet-like activity, delusions/ psychosis both agent and human, etc. It's very hard to tell, the experiment is running live. TLDR sure maybe I am "overhyping" what you see today, but I am not overhyping large networks of autonomous LLM agents in principle, that I'm pretty sure.

@paulg I hadn’t read this PG essay before. As usual, it is dense and mind-blowingly extraordinary.

Ten years ago I wrote an essay explaining the source of America's increasing polarization. If you want to understand how we got from the unity (and uniformity) of the mid 20th century to the way things are now, this is what happened: paulgraham.com/re.html

@EmmetPeppers @elonmusk @SawyerMerritt Darn - I was off by 4 days...

@EmmetPeppers @elonmusk @SawyerMerritt I’m guessing that Robotaxi without the safety monitor launches in Austin by Jan 18, but we’ll see…

NEWS: NVIDIA just announced Alpamayo, what CEO Jensen Huang calls the world’s first thinking, reasoning autonomous vehicle AI, launching on U.S. roads later this year, starting with the Mercedes CLA. Jensen: "It's trained end-to-end. Literally from camera in to actuation out; It reasons what action it is about to take, the reason by which is came about that action, and the trajectory." Alpamayo introduces Vision-Language-Action (VLA) models, which enable self-driving systems to interpret what they see, reason about complex driving scenarios, and generate driving actions. The platform includes large reasoning models, simulation tools for testing rare and edge-case scenarios, and open datasets for training and validation. NVIDIA says the approach improves transparency, safety, and robustness in autonomous systems, particularly in complex real-world environments, and supports progress toward higher levels of vehicle autonomy: "With a 10-billion-parameter architecture, Alpamayo 1 uses video input to generate trajectories alongside reasoning traces, showing the logic behind each decision. Developers can adapt Alpamayo 1 into smaller runtime models for vehicle development, or use it as a foundation for AV development tools such as reasoning-based evaluators and auto-labeling systems. Alpamayo 1 provides open model weights and open-source inferencing scripts. Future models in the family will feature larger parameter counts, more detailed reasoning capabilities, more input and output flexibility, and options for commercial usage."