Srini Rajagopal

@blader @garrytan @blader what’s your current setup? All codex or cc also in the mix?

@garrytan YES

OK Codex is GOAT at finding bugs and finding plan errors

@garrytan HBD Garry!

It's my birthday and on my birthday I want to recognize all my haters. Haters do the best marketing. Love your haters.

@pitdesi There is only one marginal manufacturer that can handle deflationary prices

Remarkable and sad that markets like this all around the world now sell Chinese goods that they try to pass off as local.

We knew the Ivy League was racist against east asian and white men, but little did we know they were EXTRA racist against south asians.

@jacobandreou @mustafasuleyman @satyanadella Congrats @jacobandreou!

Starting a new chapter today and stepping in to bring Consumer and Commercial Copilot together into one org. One team, one product, one experience. Grateful to @mustafasuleyman and @satyanadella for the opportunity. Let's build.

@Citrini7 All the best to your dad and hope there is a swift recovery

As some of you know, my dad has been fighting lung cancer for the past 2 years. If any of you have any experience with the medical system, you know that the family/patient have to be aggressive advocates. Doctors have a lot of patients and, as much as I’m sure they’d like to, don’t have infinite time. It’s been pretty amazing to see how AI has enabled my mom to be an advocate by translating medical terminology and explaining complex issues. Definitely makes things a lot easier and has made the entire process more manageable. Whenever we get a new test result, she puts it into chat GPT or Claude and (especially now that it has memory) it updates her on what’s going on. Incredibly helpful and really has maximized the communication with his team.

@rohanvarma What are the new bottlenecks?

Had a call with Cisco leadership today to check in on their Codex deployment. They’re leveraging AI coding and codex more aggressively than most startups and frontier tech companies I talk to. Skills for everything, agent swarms to execute bigger projects, automated code review, deep system integrations, and more 👀 They’ve quickly hit new bottlenecks that we’re jumping in to help them solve with more codex. If your public company exec team isn’t pushing the org to cut project timelines by 95%, maximally leverage coding agents, and reimagine the SDLC, you’re ngmi. Shoutout to Jeetu Patel - bullish on Cisco.

@jukan05 👋

Is there anyone in San Jose who’d like to have dinner with me tonight? Eating alone is a little boring…

@kchonyc @karpathy @kchonyc how did you debug this?

thanks to @karpathy , now i have cracked the mystery why my agent doesn't follow my instruction closely enough.

@JasonBud @milichab Congrats

I’m proud to be joining SpaceX and xAI with @milichab It has become clear that software is changing fundamentally. More and more, people can shape the tools they use directly, and the ceiling of what can be built keeps rising. What makes xAI special is the scale of its ambition: to build from first principles all the way out to the stars. I’m especially grateful to work on products that expand human agency and freedom. That mission is deeply personal to me. My family came to the United States fleeing communism, and the belief that freedom should be part of the next generation of the internet has driven me every day since Andrew and I started Skiff. Now, we get to work on intelligence, understanding, and freedom on a universal scale.

@altcap Doesn’t make sense then anthropic’s spend run rate would be 500B

These will prove to be among the dumbest takes of the AI era. Given how early we are & how much competition exists including free & capable models - I am shocked at the high & improving gross margins of the leading model providers. The tokens are very profitable. 🧐💰

@akshay_pachaar Good writeup! does this work for continual learning? how is reward hacking prevented?

@myprasanna Would love to try the product

I’m starting a new co in the AI coding space. Hiring the founding team members. I’ve been using the product non stop for the last month and it’s mind blowing. I think models are having different strengths right now and an ideal coding harness should run in the cloud and tap them. It should also be super simple for a lay user to configure a harness to fix common model errors you experience day to day, by throwing more model token combos at the problem. I have burnt $50k last month on this and the emergent intelligence has been epic. Lot of these we have good ideas on how to keep the intelligence and lower costs. Will release more product details soon. Join us if you’d like to work on this. Epic time for the intelligence take off and coding is driving the singularity now. What a time to be an engineer.

@ylecun @alz_zyd_ @ylecun how do you help shape this skill since the imo path is easier to measure / coach

Hugo Duminil-Copin, French mathematician and 2022 Field Medalist told me he never participated in math competition and was very bad at it. Innovative mathematics requires creativity, intuition, intense concentration, and long reflections, sometimes spread over several years. Good performance at a math olympiad merely tests fast problem solving abilities. AI can do that nowadays. One of the big activities of a researcher, in mathematics and elsewhere, is not to answer questions but to ask the right questions.

France has more Fields medalists than any other country in Europe, but performs extremely poorly on the IMO. Countries like Hungary, Romania, and Bulgaria do very well on the IMO, but don't have a single Fields medal. Why?

@TejuAdeyinka @StanfordHAI Thank you @TejuAdeyinka!

@srini @StanfordHAI Hi Srini! I heard from the team that this issue has now been resolved.

If you have a 14–16-year-old, encourage them to apply to Stanford AI4ALL (by @StanfordHAI). Incredible early exposure to AI and future careers in tech. (Applications close in 2 days) Learn more and apply: ai4all.spcs.stanford.edu

@hallerite @grok summarize what’s novel about this paper

I haven't been as excited about a paper in a long time, but MaxRL (terrible name, btw) is really elegant. The jury is still out on how this scales, but I will definitely do some ablations with this.

@nzisakiilu Thank you!

@srini I’m alumni not employee. All the best, it will prob resolve soon.

Stanford AI4ALL is open to current 9th graders (ages 14-15 during the program) interested in AI—no prior experience needed. International students can apply, with financial aid available. It offers both online (June 15-26, 2026) and residential options. Deadline: Feb 6, 2026 at 11:59pm PT. Apply here: ai4all.spcs.stanford.edu More: The Stanford AI4ALL is a 2-week program with lectures on AI topics like Computer Vision, Medical AI, NLP, and Robotics. Students do small-group research projects applying AI to societal issues (e.g., poverty detection via satellite imagery, disaster response with NLP).

@jxmnop Congrats

at long last, the final paper of my phd 🧮 Learning to Reason in 13 Parameters 🧮 we develop TinyLoRA, a new ft method. with TinyLoRA + RL, models learn well with dozens or hundreds of params example: we use only 13 parameters to train 7B Qwen model from 76 to 91% on GSM8K 🤯

@GenAI_is_real @grok what’s the oai build system that’s better than Bazel and how does it work

FAANG is literally panicking refactoring because human code is now the bottleneck. But honestly, monorepos won't save them from the infinite spaghetti code agents are about to dump. OAI already has internal tools for this that make Bazel look like a toy. The era of human "senior engineers" is ending faster than you think @karpathy @sama