Sahil 🤖

Arguing with computers for a living. Sometimes they listen.
ex Director @Twilio AI, @CapioAI (acq. Twilio), SlideShare (acq. @LinkedIn), @CarnegieMellon @mldcmu

[Justin Poehnelt @JPoehnelt]

Two months ago I was fired by Google for creating the Google Workspace CLI. It went viral, hit #1 on Hacker News, gained thousands of GitHub stars and many thousands of actual users in just a couple days.

It was an incredible, confusing journey, from directors and leaders asking what they could learn from the tool to getting grilled by legal about why the Google logo and brand colors are on the Google Workspace GitHub code repositories.

I think the cause was that Workspace and certain leaders (and projects) were afraid of being disrupted. But the fear wasn't specific to my CLI, it was a broader fear in what agents meant for Workspace. Either way, the irony of my termination was the announcement at Google Cloud Next two days before I was fired that an official Workspace CLI was coming.

I want this out there because it is easier for me to explain my story and it is an experience I want to fully own. It's also part of my healing.

Nearly 7 years at Google was an incredible opportunity for me and I was fortunate to have wonderful teammates and a manager that fully supported me through these last few months. Thank you.

[weisser @julianweisser]

Solo-founded $2B company. Used by OpenAI, Anthropic, Replit.

Solo Founders ep 13 is live with @grinich of @WorkOS.

07:49 The co-founder "blood bond" is a myth
10:55 There's only room for one Zuck at Meta
32:51 Pivots are the most traumatic thing you can do to a business
43:18 The best ideas always look like bad ideas
50:14 Sales is the ultimate expression of the company's value
53:57 The "founder pixie dust" no hire can replace
01:07:42 You only get one life. Why not swing as hard as you can?

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[Stratechery @stratechery]

Agents Over Bubbles

Agents are fundamentally changing the shape of demand for compute, both in terms of how they work and in terms of who will use them. They're so compelling that I no longer believe we're in a bubble.

https://t.co/LdPu3L37Vl

[Andrej Karpathy @karpathy]

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:

- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)

The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.

https://t.co/YCvOwwjOzF
Part code, part sci-fi, and a pinch of psychosis :)

[Claude @claudeai]

Introducing the Claude Marketplace, a way for enterprises to simplify their procurement of AI tools.

Now in limited preview.

[Julien Bek @JulienBek]

https://t.co/hsKrbn1gEq

[Thariq @trq212]

Voice mode is rolling out now in Claude Code. It’s live for ~5% of users today, and will be ramping through the coming weeks.

You'll see a note on the welcome screen once you have access. /voice to toggle it on!

[Simon Willison @simonw]

I've published the first two chapters of a new guide to Agentic Engineering Patterns - coding practices and patterns to help get the best results out of coding agents like Claude Code and OpenAI Codex https://t.co/XIskcgeBFE

[Jeff Dean @JeffDean]

Learn more about a very high quality time series model released by @GoogleResearch a while back at https://t.co/GClBKeDPhf

[Philipp Schmid @_philschmid]

August at Google DeepMind was like 🧞‍♂️ 🖼️ 🍌 🚀 🔍 🤏🏻

- Nano Banana (Gemini 2.5 Flash Image)
- Gemini Embedding
- Veo 3 Fast
- Genie 3
- Imagen 4 Fast
- Gemma 3 270M
- Perch 2
- Kaggle Game Arena
- Gemini API Url Context
- AI Studio Builder (UI Rework, Prompt Suggestions, GitHub integration …)
- AI studio UI (Model Picker, Chat, Scrolling…)

and much more!

[Paul Graham @paulg]

Something I just told a founder: Stay as small as you can for as long as you can. People who come to visit your office should always be surprised that such an important company has so few employees.

[Sahil 🤖 @SahilKumar]

@anythingai @anythingai

[Quanta Magazine @QuantaMagazine]

Earlier this year, a 17-year-old high school student named Hannah Cairo solved a 40-year-old mystery about how waves behave, surprising and exciting mathematicians. @KSHartnett reports: https://t.co/gTOeUIa9cZ

[Georg von Manstein @MansteinGeorg]

I’m a 19 y/o founder from Germany, and I cracked how Google’s Text Diffusion model works!

There are three main approaches on this topic:

[Brett Adcock @adcock_brett]

Figure is officially 3 years old 🎉

Over the last 3 years, we’ve designed three generations of robots, deployed commercially, and built the best team ever assembled

Think about what this looks like 3 years from now

[Sahil 🤖 @SahilKumar]

The new (computer use) agent in town.

[Simular @SimularAI]

🚨 New SOTA on OS-World: 41.4% success rate!

Agent S + Gemini 2.5 Pro outperforms the previous SOTA held by Agent S + Claude 3.7 (34.5%) — once again, we beat ourselves.

And we’re now significantly ahead of OpenAI (32.6%) and Anthropic (26%). 🔽

[Andrej Karpathy @karpathy]

This is interesting as a first large diffusion-based LLM.

Most of the LLMs you've been seeing are ~clones as far as the core modeling approach goes. They're all trained "autoregressively", i.e. predicting tokens from left to right. Diffusion is different - it doesn't go left to right, but all at once. You start with noise and gradually denoise into a token stream.

Most of the image / video generation AI tools actually work this way and use Diffusion, not Autoregression. It's only text (and sometimes audio!) that have resisted. So it's been a bit of a mystery to me and many others why, for some reason, text prefers Autoregression, but images/videos prefer Diffusion. This turns out to be a fairly deep rabbit hole that has to do with the distribution of information and noise and our own perception of them, in these domains. If you look close enough, a lot of interesting connections emerge between the two as well.

All that to say that this model has the potential to be different, and possibly showcase new, unique psychology, or new strengths and weaknesses. I encourage people to try it out!

[Andrej Karpathy @karpathy]

New 3h31m video on YouTube:
"Deep Dive into LLMs like ChatGPT"

This is a general audience deep dive into the Large Language Model (LLM) AI technology that powers ChatGPT and related products. It is covers the full training stack of how the models are developed, along with mental models of how to think about their "psychology", and how to get the best use them in practical applications.

We cover all the major stages:
1. pretraining: data, tokenization, Transformer neural network I/O and internals, inference, GPT-2 training example, Llama 3.1 base inference examples
2. supervised finetuning: conversations data, "LLM Psychology": hallucinations, tool use, knowledge/working memory, knowledge of self, models need tokens to think, spelling, jagged intelligence
3. reinforcement learning: practice makes perfect, DeepSeek-R1, AlphaGo, RLHF.

I designed this video for the "general audience" track of my videos, which I believe are accessible to most people, even without technical background. It should give you an intuitive understanding of the full training pipeline of LLMs like ChatGPT, with many examples along the way, and maybe some ways of thinking around current capabilities, where we are, and what's coming.

(Also, I have one "Intro to LLMs" video already from ~year ago, but that is just a re-recording of a random talk, so I wanted to loop around and do a lot more comprehensive version of this topic. They can still be combined, as the talk goes a lot deeper into other topics, e.g. LLM OS and LLM Security)

Hope it's fun & useful!
https://t.co/75mXcUBI8L

[Verbis @VerbisAI]

🚀 Thrilled to launch Verbis(https://t.co/s4KdnEe8Gt), an open-source MacOS app leveraging GenAI for secure, local data processing. Your workplace search assistant now on your own machine! 🛡️

GitHub: https://t.co/5a59HXPs7a
Discord: https://t.co/AVyMAbtEy0

#GenAI #OpenSource