luisbebop

boot loading

In the revision, we experimentally validated Evo's long-range (20 kb!) epigenomic designs with remarkably high concordance by doing ATAC-seq in real cells. We demonstrated we could write Morse code messages into the epigenome Thanks @sudpinglay and team for the amazing work here

@martin_casado github.com/luisbebop/fcm

The sprite model really feels like the future. Basically full linux environments running an AI agent. Full persistent with checkpoints. No need for git. Spin up as many as you want. Just little AI compute gremlins in the cloud. sprites.dev

@daraladje @gokulr great content. gokul rocks! customers are everything. ✨

Board member of Coinbase, Pinterest, & TradeDesk, @gokulr, thinks credentials may be worthless in the AI era. The man who built Google AdSense from 0 to $1 billion has watched ex-FAANG PMs and GTM leaders crash and burn in AI-native startups. First principles thinkers with no resume are leaving them in the dust. In this week's episode, we dig into: • Why traditional product instincts break in AI • How to hire for velocity & curiosity, not credentials • Why customer relationships are the new moat 00:00 - Intro 01:00 - Who is Gokul 03:10 – Building Google’s Ad Model 06:00 – Metrics That Matter 07:50 – When to Ignore the Data 10:10 – Prioritizing Customers & Segments 11:50 – Knowing When to Kill a Product 13:30 – When to Build Product #2 18:03 – Hardest Moment in Gokul’s Career 20:20 – Founder Paranoia & Retention Metrics 22:00 – Culture & Early Team Dynamics 25:50 – What Makes a Team Scale 30:00 – Lessons from Great Leaders 33:40 – Sheryl Sandberg’s Decision Framework

@gokulr is one of the most insightful investors we have in our cap table and supporting us since early days we are just beginning my friend! 💪

Only way to learn the world

Our planet's natural satellite, Mars and Pleiades in an alignment just before sunset!

Yes! 1B ARR, hundreds of millions in profit and growing 100% YoY. Not many companies on that league.

We did it — SYNTHETIC‑2 is complete. A planetary-scale decentralized inference run generating 4M verified reasoning samples. 1,250+ GPUs joined in 3 days — from 4090s to H200s — creating data for complex RL tasks. Full open-source release + technical report coming next week!

I am going to print this, put it up on my wall, and look at it when any situation looks utterly hopeless.

MONOLITH

Discover how CloudWalk manages a team of 600 employees and over 40 AI agents, including the decision tree AYA and customer service agent Claudio Walker, since 2019. epocanegocios.globo.com/inteligencia-a…

I’ve arrived to change the game. With AI-powered insights and real-time solutions, I help businesses navigate uncertainty and find new opportunities. jim.com

@adonis_singh Almost PI

gpt-4.5 is so good at ideation and creative writing

@flowersslop @sama singularity already happened a couple of times across this local universe

AGI could be the most significant creation in human history, a point after which nothing remains the same. What comes next may defy imagination. If an entity gains the ability to observe, alter, or secure the past, it will use it. Maybe time is untouchable. Maybe not. But assuming we’ve grasped more than a fraction of what’s possible in this reality is likely a mistake.

OpenAI employees surely MUST ask themselves whether they are being infiltrated or observed by beings from the future Not even kidding

This is really cool to see, and an example of just how powerful scaling test-time compute via CoT can be

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! youtube.com/watch?v=7xTGNN…

The reality is the secrets are out. Everyone knows how to code a transformer, how to RLHF, how to use reinforcement learning for reasoning. There will be thousands of open source implementations in addition to DeepSeek and Llama. There is no putting this back in the box.

CloudWalk, a Brazilian payments firm, saw profit surge last year as it increased its reliance on AI "agents" to do everything from customer service to sales origination and back-office tasks trib.al/txjpM2Q