Sumanth Dathathri

@reasoningmodels @SawyerMerritt Extending the same line of reasoning, Apple & Google should have a valuation of 14 T (they're both around ~4.5x smaller)

@SawyerMerritt Pretty incredible, generating $2B/mo, so $24B/year. So 35x annual revenue multiple. This is like a company doing $10M/year in ARR raising at a $350M valuation. Which I think does happen quite a bit, so maybe nothing too unusual here right??

NEWS: OpenAI just announced that it has officially closed their latest funding round with $122 billion in committed capital at a post money valuation of $852 billion. "We are now generating $2B in revenue per month. At this stage, we are growing revenue four times faster than the companies who defined the Internet and mobile eras, including Alphabet and Meta. ChatGPT has more than 900 million weekly active users, and over 50 million subscribers. Search usage has nearly tripled in a year, and our ads pilot reached more than $100 million in ARR in under six weeks. Momentum is just as strong on the enterprise side, which now makes up more than 40% of our revenue, and is on track to reach parity with consumer by the end of 2026. GPT‑5.4 is driving record engagement across agentic workflows. Our APIs now process more than 15 billion tokens per minute. Codex now serves over 2 million weekly users, up 5x in the past three months, with usage growing more than 70% month over month."

@pfau TBF, Alec is an excellent name for a chatbot!

Oh god are we really doing this? Jeff Dean trained an n-gram model on the entire internet in 2007. Jelinek coined the term "language model" in the '70s. It's called "Claude" because Claude Shannon was estimating the entropy rate of the English language in 1951!

@deedydas Very good engagement farming!

Google Senior Staff Engineer to me: “Yeah, I have no clue what Claude Code / Codex is but I hear it’s all the rage. No, I don’t really care, I just need GOOG to hit $400 and keep this job for 2-3 more years so I can retire!”

@sedielem @volokuleshov I understand Elon's tweet as implying it's the modeling paradigm that all the hype should be about, not the architecture. He knows it all.

@volokuleshov What does he know? Not the difference between a model architecture and a modelling paradigm, apparently🙃

Elon knows.

@srodan Don't know if the Bengals famine victims fully agree.

This is what so-called « pro-Palestinians » did last night to the man who saved the free world.

I'd like to thank @daniel_rossett for his help in my recovery from the POTS version of Long COVID. Daniel was key in bringing me back from highly disabled and suffering to being able to do what I want to again. This X account is mostly focused on ML / AI. From that point of view, many of you know that in December 2024, I wasn't able to do the test of time award talk at NeurIPS, even by video call. Daniel started working with me in March 2025. By April, I started to have days of no POTS symptoms, by June I was off all heart rate lowering medications, by September I was back to work. I'm back to full exercise, running, lifting weights, mountain biking, and have even done things I hadn't done before I got sick, like riding Whistler Mountain Bike Park. I'm now getting the word out to help Daniel build a company that will bring this approach to more people.

@sunny2dutta @original_ngv "You cannot chose any major you want. You have to take appropriate courses say Intro to Programming, math courses if you do well there you can chose a CS major"-- this isn't true. You need to make it through a set of core courses, that set of courses is the same for all majors.

1. You cannot chose any major you want. You have to take appropriate courses say Intro to Programming, math courses if you do well there you can chose a CS major. Many advanced electives have limits. IIT equivalent would be branch change, you do very well in year 1 (GPA 9+) you can chose any major you want. 2. Also different universities operate differently Oxford, Cambridge you are supposed to chose your field right in the beginning some are more competitive than others.

Then why do Ivy Leagues let students choose their own majors? Surley a university like MIT or Harvard is smarter than IIT, no?

@__apf__ @devahaz @rohindhar As someone who lives in central-ish London, 2000ft is like the Buckingham Palace.

@devahaz @rohindhar any house in the lower hills (meaning it has views but isn't in the high risk fire zone) is going for crazy amounts. check out this one. 1.2M in 2019, 2.2M a few months later, 2.6 in 2024, and 3.4 in 2025. only 2000 sqft. zillow.com/homedetails/19…

San Francisco home sale in Pacific Heights (not necessarily the fancy part) Listed for $5.995MM Just sold for $8MM

@NirantK Would be very happy to sell you USD for 95 INR sir.

PSA: If you're doing USD to INR conversion mentally, can just use 100 now.

@iliaishacked That's already what's happening with self-driving. The perception systems are just far from robust.

west coast: we are weeks from the singularity east coast: what will this mean for saas multiples? also, somehow, tacos are ableist melbourne, australia: 22c/72f and sunny, nice breeze, might take my family to the zoo

Folks, we are hiring a few systems researchers/engineers (full-time, part-time or internships) with the following requirements: * Experience in systems research * Familiarity with inference stacks such as vLLM, SGL, or TensorRT * Python, CUDA, and Rust experience is a strong bonus

@iScienceLuvr Inb4 automated labs.

Honestly, I expect AI researchers to get replaced by AI agents before other researchers because most other research disciplines require physical experimentation.

@vishalmisra Where's the spike once you've had a kid?

Most depressing chart I have seen in years :-(

I started my career as a theorist, and am now an empirical LLM researcher. In today's blog post, I talk about the parallels between theory and empirical research: kamalikachaudhuri.substack.com/p/a-theorists-…

@SawyerMerritt @Tesla How do bats and dolphins navigate their environments? Clearly with their eyes.

Ashok Elluswamy, VP of AI at @Tesla on self-driving: "It's so obvious you can solve this with cameras. Why wouldn't you solve with cameras? It's 2026. The self-driving problem is not a sensor problem, it's an AI problem. The cameras have enough information already. It's a problem of extracting the information, which is an AI problem." (via @aelluswamy's presentation at the 2026 ScaledML Conference on January 29th)

@iliaishacked @MLStreetTalk does nothing

@MLStreetTalk adversarial training

writing is the adversarial process with yourself to become more coherent

@geoffreyirving We may actually have a use case for this! Very useful!

One of the more useless things I did while at Google Brain was write down random access into xorshift128+, the hardware random number generator on TPUs. Purely a stunt: it could theoretically have meant TPU-native Jax-style random numbers faster than Threefry, but in practice random numbers are cheap and the complexity is not worth it. I still have the code in a branch, but certainly no one has ever used it. Fun way to learn about finite field isomorphisms, though. github.com/tensorflow/ten…

@kaytraser Not sure I follow. E.g., the KL regularization does push the model towards pretraining, so it should increase the influence of pretraining?

@sdathath this seems to be more aligned with the task of pure next token generation hence the suspension of being more influenced by changes in pre training

did synth data generation for the same task in Sept 2024 and today fighting mode collapse was so hard back then and is completely absent now we've came a long way, wondering if it is only because models got larger or did the labs actually get an improved data distribution

@kaytraser I think RL inherently isn't mean to take away all entropy, the way it's been doing for a while. I think it's a sign people are getting better at RL with LLMs.

@kaytraser You mean worse diversity? It depends on your KL regularization wrt original LLM dist. If you have very strong reg, you'll stay close to the original distribution. I've seen things trending towards a bit more care around stopping the model from collapsing these days.