We formally announced Mirendil less than 4 months ago. Today, I'm excited to share an important milestone for: Pretraining at scale without a pretraining team. Mirendil's mission is to solve Recursive Self-Improvement (RSI) and use it to democratize frontier AI development, ultimately accelerating science and technology. At Mirendil, when we think about RSI, we think holistically about the company as an intelligent entity that self-improves toward its goal. From that perspective, the most powerful notion of acceleration is reducing the time and resources (number of human experts, compute, and other costs) needed to achieve ambitious goals. Given that pretraining is perhaps the most important yet most challenging part of frontier AI development, we decided to run an ambitious controlled experiment: Starting fully from scratch, can we significantly reduce the time, number of human experts, and compute resources needed to go all the way to successfully pretraining a strong coding model at scale? Mirendil-Base is the result of this experiment. Our first pretrained model was built by one researcher in 3.5 months, with less than $10M in compute. This includes all the research and small-scale experiments it took to get there, as well as the final training run. The recipe holds up as the models get bigger. For reference, Mirendil-Base is in the 100B-500B parameter range, Mirendil-Base-Large in the 500B-2T range, and Mirendil-Base-XL in the 2T-10T range. We're beginning to see a world where a single researcher, empowered by AI, can take on work that once required an entire research organization. As these systems improve, the bottleneck shifts from the size of a team to the ambition and judgment of the people directing them. This has profound implications for who gets to build AI and, ultimately, who gets to push the frontiers of science and technology forward. We’ll share independent evaluations, post-training benchmarks and API-access sooner than you think so stay tuned!

I made a prompt that sorts your inbox every morning before you open it, using Dwight Eisenhower's urgent and important box. In August 1954, Eisenhower told an audience at Northwestern University: "I have two kinds of problems, the urgent and the important." He credited the line to a former college president. Productivity writers later turned it into a 4-box grid and put his name on it. Paste this into Claude, or set it up as a Grok Bot routine: --- You are running an Eisenhower box on my messages every weekday at [TIME] in [TIME ZONE]. Sources: my Gmail and my X Chat messages from the last 24 hours. What matters to me this month: [YOUR 2 OR 3 PRIORITIES] Sort every message into 4 boxes: 1. Urgent and important: needs me today 2. Important, not urgent: moves a priority forward 3. Urgent, not important: someone else's deadline 4. Neither: safe to archive Then send me: 1. Box 1, with a draft reply for each 2. Box 2, with a suggested day this week for each 3. Box 3, with who could handle it or a 1-line reply 4. A count for box 4 Rules: if you're unsure which box, ask me one question at a time. Don't send, archive or delete anything until I approve. --- Bookmark this for the next time you open 80 unread messages before breakfast.

what did you get done with grok @bot last week?

after 300+ hours in Grok Bot, i can tell you most people are using it WRONG they treat it like a chatbot... i have 20 bots working while i sleep here's the full guide to building your own AI agent team: 📁 the mistake ↳ most people build one giant bot and ask it everything ↳ or they just chat with it like a search box ↳ one bot doing every job gets messy fast... it mixes up context and does everything half right 📁 one bot, one job ↳ every bot gets exactly one job ↳ a research bot only researches ↳ a writing bot only writes ↳ small, focused bots are easier to fix and easier to trust 📁 your first bot ↳ give it a clear job in one line ↳ give it a tone ↳ how it talks, how long its answers are ↳ tell it where your stuff lives ↳ which docs, folders and accounts it should look at ↳ correct it once and it remembers ↳ you stop repeating yourself every morning 📁 connectors ↳ plug your bots into the tools you already use: ↳ Notion for notes and docs ↳ Gmail for email ↳ GitHub for code ↳ Buffer for scheduling posts ↳ X for research ↳ now they work on your real stuff, no copy paste 📁 chief of staff ↳ you only talk to ONE bot, the chief of staff ↳ it does no specialist work itself... it just routes ↳ results come back to you through it 📁 the org chart ↳ under the chief of staff sit the specialists: ↳ X research bot ↳ Reddit research bot ↳ YouTube lessons bot ↳ site audit bot ↳ copywriting bot ↳ bots message each other ↳ the research bot hands its findings straight to the copywriting bot, no you in the middle 📁 bot templates ↳ Grok Bot has bot templates you can start from instead of building from scratch ↳ or share one with a link ↳ Alfred, my bot chief advisor, designs, audits and improves my other bots ↳ dr eggcbot asks a few questions, then creates new bots for me ↳ bots build bots... bots fix bots 📁 routines ↳ recurring work on a schedule that runs while you're offline ↳ examples: ↳ every morning the X research bot pulls what's trending in your niche ↳ every week the site audit bot checks your site and sends a report ↳ every night the Reddit bot collects the questions people ask in your space ↳ you wake up to work already done 📁 skills ↳ any task you repeat, save it once as a skill ↳ every bot on the team can use it ↳ write the steps once, never explain them again 📁 give it Claude Code ↳ tell your bot to install Claude Code in its terminal ↳ log in ↳ now it builds its own scripts and tools ↳ need a scraper or a small helper tool? it builds it 📁 style files ↳ have a bot study a creator's top posts ↳ it writes a tone .md file ↳ voice, hooks, line rhythm, the words they use ↳ every writing bot reads tone .md first ↳ no more generic ai slop 📁 sharing passwords with 1Password ↳ first create a separate vault just for Grok Bot (call it "Grok Bot") ↳ put only the logins your bots need in it, nothing else ↳ connect that vault, not your whole 1Password ↳ logins stay saved and encrypted in 1Password, bots only get what you put there ↳ at a login page the bot fills the saved login ↳ it never sees the actual password... it only knows if the fill worked ↳ no pasting passwords in chat 📁 safety ↳ bots draft, you press send ↳ they ask before deleting, buying or posting anything ↳ the bots do the work, you keep the final say pro tip: copy this tweet into Grok Bot and see what team it builds around YOUR workflow stuck somewhere? drop it below, i'll help where i can

I just created a Grok Bot 101 using my real use case. (Using Grok Bot + Grok Voice) - Made a Grok Voice clone of me - Gave my script to the voice - Dropped the voice clip into Grok Bot - Grok Bot made the video with captions and animated visuals