Vikram Rawat

NVIDIA CEO Jensen Huang: “I really discourage 1-on-1s” Jensen famously has 60 direct reports. When Stripe founder Patrick Collison points out that this isn’t conventionally considered best practice, Jensen shares his reasoning: “I don’t do 1-on-1s, and almost everything I say, I say to everybody all the time. I don’t really believe there’s any information that I operate on that only one or two people should hear about… I believe that when you give everybody equal access to information, that empowers people. And so that’s number one… Number two, if the CEO’s direct staff is 60 people, the number of layers you’ve removed in a company is probably something like seven.” Patrick offers to steal man the other side of the argument: “1-on-1s are where you provide coaching, where you maybe talk through personal goals and career advancement, where maybe you give feedback on something that you see somebody systematically not doing so well… Do you not do those things or do you do them in a different way?” Jensen responds: “I give you feedback right there in front of everybody. In fact, this is a really big deal. First of all, feedback is learning. For what reason are you the only person who should learn this?… We should all learn from that opportunity… Half the time I’m not right, but for me to reason through it in front of everybody helps everybody learn how to reason through it. The problem I have with 1-on-1s and taking feedback aside is you deprive a whole bunch of people that same learning. Learning from other people’s mistakes is the best way to learn.” Video source: @stripe (2024)

Amazon is holding a mandatory meeting about AI breaking its systems. The official framing is "part of normal business." The briefing note describes a trend of incidents with "high blast radius" caused by "Gen-AI assisted changes" for which "best practices and safeguards are not yet fully established." Translation to human language: we gave AI to engineers and things keep breaking? The response for now? Junior and mid-level engineers can no longer push AI-assisted code without a senior signing off. AWS spent 13 hours recovering after its own AI coding tool, asked to make some changes, decided instead to delete and recreate the environment (the software equivalent of fixing a leaky tap by knocking down the wall). Amazon called that an "extremely limited event" (the affected tool served customers in mainland China).

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. github.com/karpathy/nanoc… All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

Just a reminder why electric vehicles are better than gas cars:

I’m American. After my PhD, I went to India. What I experienced dismantled my Western worldview. Here are 8 lessons that permanently rewired how I see life:

This is the beauty of calculus. Wish someone had taught it like this to me. ( 🎥 Dm fro credit)

A Brazilian farmer playfully calls out each cow’s name from his notebook, and the cows answer with distinct moos proof that daily routines have taught them to recognize their names.

🚨 BREAKING: In a stunning blow to Democrats, 76% PERCENT of BLACK Americans want nationwide voter ID — in other words, the SAVE America Act White voters: 85% want it Latino voters: 82% want it Another leftist narrative just got decimated. Pass voter ID. GET THIS PASSED. 🇺🇸

Just Recorded a 1-hour FULL Course on how to create STUNNING Animated websites with AI ❤️‍🔥You can literally make ANY Design with AI

ClawdBot/Moltbot/Openclaw's CEO on the moment he realized he'd accidentally built the world's first real Jarvis.

Instead of watching a 2-hour movie, watch this Claude FULL COURSE (Build & Automate Anything):

The @pmarca episode Marc is a many-time founder, investor, co-creator of the web browser, co-founder of Netscape, the "a" in @a16z. In this very special conversation, we dig into why we’re living through one of the most unique times in history, and what comes next. We discuss: 🔸 Why AI is arriving at the perfect moment to counter demographic collapse and declining productivity 🔸 How Marc is raising his 10yo kid to thrive in the AI future 🔸 The 3-way standoff that’s happening between PMs, designers, and engineers 🔸 What’s actually going to happen with AI and jobs (spoiler: he thinks the panic is “totally off base”) 🔸Why AI is "the philosopher’s stone" made real 🔸 Why you should still learn to code 🔸 His media diet: X and old books, nothing in between Don't miss this one. Full episode embedded here on X 👇

I'm the VP of AI at Apple. I've been here since 2011. I watched Siri launch. It was revolutionary. For about six months. Then Google Assistant came out. Then Alexa. Then ChatGPT. We kept saying we were "focused on privacy." Privacy is what you say when you're losing. Three years ago the board asked about our AI strategy. I showed them a slide that said "On-Device Intelligence." They nodded. They didn't know what it meant. Neither did I. But it had a picture of a neural network. Neural networks look impressive. Even when they don't work. Last year someone asked Siri to set a timer. It opened a Wikipedia article about timers. Tim saw the meme. He didn't laugh. He scheduled a "strategic offsite." Offsites are where we go to admit failure privately. I presented three options. Option 1: Build our own LLM. That would take four years. We don't have four years. Option 2: Buy a startup. We looked at twelve. They all wanted $40 billion. For teams of nine people. Who would leave after the acquisition. Option 3: Call Google. The room went quiet. Google is the enemy. We've spent fifteen years pretending we're better than Google. Our entire brand is "not Google." But Google has TPUs. We don't. Google has Gemini. We have Siri. Siri still can't reliably add items to a grocery list. I called Sundar. He picked up on the first ring. He'd been waiting. They all wait. Eventually everyone calls Google. I asked for TPU access. He said yes. I asked for Gemini integration. He said yes. I asked how much. He said one billion dollars. I said that's a lot. He said "per year." I paused. He said "you don't really have a choice." He was smiling. I could hear it. We announced it as a "strategic partnership." Partnership means we're paying them. The press release said we're "enhancing Siri's capabilities." Enhancing means replacing. We said the new Siri arrives "late 2026." Late 2026 means 2027. Maybe 2028. Definitely not 2026. A reporter asked if this means Apple lost the AI race. Our comms team said we're "thoughtfully deliberate." That's not an answer. But it has enough syllables to sound like one. Internally, we're calling it "Project Humble Pie." Someone suggested "Project Brain Transplant." HR flagged that as "not brand-aligned." The engineers are relieved. They've been trying to make Siri work for years. Now they can blame Google. Blame is a renewable resource. Tim did a podcast. He said AI is "a profound technology." He's never used ChatGPT. I showed him once. He asked why it was typing so slowly. I said that's how it works. He said "Siri should be faster." I said "Siri will be Google." He said "don't say that publicly." I won't. Publicly, we're "leveraging industry partnerships." Leveraging means surrendering. But with dignity. We still have the best hardware. We still have the ecosystem. We still have the brand. We just don't have AI. So we're renting it. From the company we've mocked for two decades. The one billion dollars is a licensing fee. The real cost is the narrative. We were the innovators. Now we're the integrators. But the stock is up 3%. Wall Street doesn't care about innovation. Wall Street cares about not falling behind. We're not falling behind anymore. We're being carried. By Google. For one billion dollars a year. I'll present this as a win at the next all-hands. Wins are whatever you frame them as. The graph will go up and to the right. It always does. As long as you pick the right metric.

Apple CEO Tim Cook has a counterintuitive philosophy for achieving excellence: stop following the rules entirely. "I think you should rarely follow the rules. I think you should write the rules." When asked about when it's okay to break the rules, @tim_cook reframes the question itself as flawed. He explains: "If you do follow things in a formulaic manner, you will wind up at best being the same as everybody else. Maybe you missed something and you're a little worse. And if you want to excel, you can't do that." Tim has seen this play out repeatedly in business: "I've watched a lot of companies do that. And I think that's a rotten strategy. Maybe they'll be good for a few months or something." But he's careful to distinguish rule-writing from ignorance. The real skill, he argues, is mastering: "How to learn, how to collaborate, how to think about something, and how to work with people that have a very different point of view than you do." In other words, master the fundamentals of thinking and collaboration —then use those skills to author your own playbook rather than following someone else's. His conclusion is direct: "That's what I really think about it. Write your own rules."

There’s Tesla and then there’s every other car brand. We are not built the same.

Jensen Huang is on fire at CES 2026 Event 🎯 "How you run the software, how you develop the software fundamentally changed. The entire five-layer stack of the computer industry is being reinvented. You no longer program the software, you train the software. You don't run it on CPUs, you run it on GPUs. And whereas applications were pre-recorded, pre-compiled, and run on your device, now applications understand the context and generate every single pixel, every single token, completely from scratch every single time. Computing has been fundamentally reshaped as a result of accelerated computing, as a result of artificial intelligence. Every single layer of that five-layer cake is now being reinvented. Well, what that means is some ten trillion dollars or so of the last decade of computing is now being modernized to this new way of doing computing. What that means is hundreds of billions of dollars, a couple of hundred billion dollars in VC funding each year, is going into modernize and inventing this new world. And what it means is a hundred trillion dollars of industry, several percent of which is R&D budget, is shifting over to artificial intelligence. People ask where is the money coming from? That's where the money is coming from. The modernization to AI, the shifting of R&D budgets from classical methods to now artificial intelligence methods. Enormous amounts of investments coming into this industry, which explains why we're so busy. And this last year was no difference. This last year was incredible. This last year... there's a slide coming."

Holy Moly, Elon Musk: AI can already automate ~50% of white-collar tasks. Bits before atoms. Offices fall first, factories later. Inertia is the only delay. But blue collar workers are next. Its only a matter of (very short) time!