Daily Brief: Community-Owned Action Model AI Surpasses 100,000 Users One Week After Invite-Only Launch

Community-Owned Action Model AI Surpasses 100,000 Users One Week After Invite-Only Launch

Action Model, billed as the world's first community-owned Large Action Model (LAM), hit 100,000 users just one week after its invite-only launch around February 10.[1][2][3] Users contribute data to train the model and earn $LAM tokens for governance and ownership stakes, positioning it as a direct challenge to Big Tech's centralized AI empires.

The decentralized approach has sparked excitement, with community members calling it a "community-owned alternative that changes how AI gets built," as @Fadhil_Gamers put it on X.[1] Thousands of views and hundreds of reposts underscore the buzz around this shift toward user-driven AI development.

Andrej Karpathy Releases microGPT: Minimal LLM Trained from Scratch in 243 Lines of Python

AI luminary Andrej Karpathy dropped microGPT on February 12—a bare-bones LLM trained and run in just 243 lines of pure Python, no dependencies needed.[4][6] Building on his nanoGPT, it uses micrograd for backprop to demystify LLMs for learners, making the black box of training accessible to anyone with a laptop.

The release has educators and tinkerers buzzing, with 205 replies on Karpathy's X post praising its simplicity for teaching core concepts.[4] As @karpathy himself noted, it's an "art project" to strip AI down to essentials.

Mistral AI's Annualized Revenues Exceed $400 Million, Up 20x Year-Over-Year

French AI powerhouse Mistral AI smashed past $400 million in annualized revenue as of February 11—a staggering 20x jump from $20 million last year, according to CEO Arthur Mensch.[7][8][9] This surge rides Europe's hunger for homegrown models, dodging reliance on U.S. giants.

X chatter highlights Europe's AI momentum, with Seb Johnson on LinkedIn calling out the "20x'ed revenue... expanding RAPIDLY."[7] It's proof positive that open-weight models can scale commercially without the full closed-source baggage.

Meta FAIR Launches AIRS-Bench: Benchmark for Autonomous AI Research Agents

Meta's FAIR team unveiled AIRS-Bench on February 6, a tough new benchmark with 20 ML tasks to test AI agents' end-to-end research chops.[10][11] Agents get 24 hours and a GPU to beat state-of-the-art papers, pushing boundaries in autonomous science.

AI researchers are diving in, as @j_foerst announced on X: "Happy to release AIRS-Bench... test autonomous ML abilities."[10] This could redefine how we measure agentic AI beyond chat.

Mistral AI Commits $1.4 Billion to Build AI Data Centers in Sweden

Mistral is doubling down on Europe with a €1.2 billion ($1.4B) pledge for a 23MW AI data center in Borlänge, Sweden, at EcoDataCenter's site—announced February 11.[13][14][15] It's a bold infrastructure play for AI sovereignty amid U.S. dominance.

Nordic watchers on X hail it as a "major push into European AI infrastructure," per @GBAFReview, amplifying Sweden's role in the region's compute race.[13]

Station F Unveils F/ai Program to Accelerate AI-Native Startups in Europe

Paris-based Station F, Europe's biggest startup campus, launched F/ai on February 11 to supercharge AI-native ventures.[16] Backed by Cloudflare, OVHcloud, Snowflake, Qualcomm, and more, it aims to propel founders to global scale.

Developer communities are engaging heavily, seeing it as vital support for Europe's AI startup ecosystem.[16]