🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Happy to see that the The White House AI Action Plan includes many of our ARC Prize recommendations and prioritizes the values we push everyday: - Open source - Accelerating AGI research - Creating standard US-led AI benchmarks - Cultivating a deeper understanding of, and more

Andrej Karpathy explains Elon Musk’s unique management style

Right before #imo2025, together with colleagues from Mountain View, NYC, Singapore, etc, we all gathered at Google DeepMind headquarter in London for our final push for IMO. I believe that week was when all magic happened! We put all individual recipes (that we figured out

Slides for my lecture “LLM Reasoning” at Stanford CS 25: dennyzhou.github.io/LLM-Reasoning-… Key points: 1. Reasoning in LLMs simply means generating a sequence of intermediate tokens before producing the final answer. Whether this resembles human reasoning is irrelevant. The crucial

I'm observing a mini Moravec's paradox within robotics: gymnastics that are difficult for humans are much easier for robots than "unsexy" tasks like cooking, cleaning, and assembling. It leads to a cognitive dissonance for people outside the field, "so, robots can parkour &

Getting these examples of AI efficiency nearly every day from customers right now. 60X faster workflows due to AI Agents being able to comb through unstructured data. So many annoying tasks in knowledge work are just going to disappear because of agents.

This! Plus an even smaller fraction can appreciate how models are getting better since that requires deep expertise.

xAI commissioned us to analyze Grok 4’s math capabilities. Our findings: + It’s good at involved computations, improving at proofs (from a low base), and useful for literature search. - It favors low-level grinds and leans on background knowledge. Read on for examples!

The difference between people getting extreme leverage from AI Agents vs. moderate returns seems to be just a willingness to put the time into really good prompt design and fixing any errors from the agent. Once you make this mental jump, the output goes up enormously.

every moving machine will require an inference chip, and the capability of that machine will be governed by the integrated capability of that chip

Incredible scientific result! gilead.com/news/news-deta…

SpaceX does more meaningful, cutting-edge “research” on the advancement of rockets and satellites than all the academic university labs on Earth combined. But we don’t use the pretentious, low-accountability term “researcher”. Engineer.

This is the way!

Here's a F.02 in my home, using Helix to do my laundry

progress on hallucinations:

It’s time to say goodbye to the days of out-of-policy travel spending, approval ping-pong, and uploading crumpled receipts when traveling for business. Introducing Rippling Travel—live today. We’ve made it insanely easy for employees to book in-policy flights, hotels, and cars,

In other tech news - Rippling is launching corporate travel. - Control travel spend before it happens ("only allow booking flights within 30% of the cost of cheapest reasonable flight"). - Because it's integrated with Rippling, you can easily build out

Did anyone try using models to write Pandas to solve BIRD-bench? (bird-bench.github.io) I wonder if the performance is similar to Text-to-SQL or better.

Our IMO journey continues: the yolo run model that we trained a week before #imo2025, despite all possible likelihood of failures, magically achieves SOTA across a wide range of reasoning tasks from maths, to coding, and challenging knowledge. I'm very excited that we have now

If you're building an AI agent to automate some job, like a payroll specialist, the hardcore move is to just go get that job and do it for a while. We call this "going undercover", and we're seeing more of the top founders do it. It's the best way to learn what to build.