! TOP NEWS !
Anthropic just released Claude Fable 5, their first public Mythos-class model. SOTA on nearly every benchmark, free until June 22, and already sparking a fight over hidden safety steering 🔥
Microsoft's VS Code extensions got hit. Hackers injected malware to steal GitHub tokens — go check your environment 🔒
OpenAI filed a confidential S-1 — $850B valuation, no firm timeline, and they published it themselves before it could leak. Anthropic filed two weeks ago, SpaceX IPO is next week 📈
! PRODUCT NEWS !
OpenAI also published their "third phase" vision — automate research, give everyone a personal AGI, and a proposal to pause frontier AI globally 📜
Cognition released a benchmark where top AI coding models score 13 out of 100 on code real maintainers would merge 🧪
NotebookLM now has a sandboxed computer inside each notebook, can crawl its own sources, and exports to PDF, XLSX, PPTX 📓
Kimi just launched a desktop agent that runs up to 300 parallel AI workers locally 🖥️
! FOR THE CURIOUS !
👀 Anthropic just dropped the best model in the world. There's a catch.
So Anthropic dropped Fable 5 on Monday. The benchmarks are insane. 80.3% on SWE-Bench Pro, 11 points ahead of anything else out there. Karpathy tried it and said it's the kind of model where you just give it the hard stuff and walk away. That good.
Then people read the fine print.
Turns out Fable 5 has silent safeguards baked in. It just quietly gets worse at certain things through prompt modification or steering vectors, and it doesn't tell you when it's happening. Anthropic says it affects 0.03% of traffic. But the open-source community lost it. Jeremy Howard called it "a very dark and very sad day." Others flagged antitrust risk. Someone called it fraud.
Best model ever shipped. You just can't tell when it's sandbagging.
🚀 Your agents can now generate any dataset on-the-fly
Your agents can reason, plan, and act, but they're still starving for clean structured data. BigSet fixes that.
Describe any dataset in a sentence, like "AI startups hiring ML engineers, with funding stage and headcount," and agents build it from the live web, checking every row against real sources. No scrapers, pipelines, or configs.
GPU prices, who's hiring, competitor features, whatever you type. Set a refresh cadence from 30 minutes to weekly and it stays current. Every row ships with source links. Runs as a CLI so Claude Code, Codex, or OpenClaw can build datasets themselves.
npm install -g @adamexu/bigset
! BANGER EVENTS !
Agentic Engineering Summit - SF
Founders, investors, technical leaders from the best companies on what's working, what's breaking, and what's next.
🗓️ Thurs, June 18 . 9:30 AM PT
Multiagents Hackathon - LDN
Build, experiment, and ship alongside fellow builders with Google DeepMind & more.
🏆 $20k+ in Prizes!
🗓️ Fri, June 26 . 9:30 AM BST
Afterglow Summit Afterparty - SF
Grab a drink, meet the people you missed, continue conversations and keep the night going.
🗓️ Thur, June 18 . 6:30 PM PT
Agents & Bagels: Breakfast meet up
A group of agent builders and founders creating production-grade agents for real customers meets informally before work. Free, casual, and surprisingly useful.
🗓️ Wed, June 17 . 8:00 AM PT
Interested in sponsoring an upcoming event or hosting your own?
reach out here
NotebookLM now has a sandboxed computer inside each notebook — can crawl its own sources and export to PDF, XLSX, PPTX
Kimi launched a desktop agent that runs up to 300 parallel AI workers locally — from the same team reportedly raising at a $30B valuation
DeepSeek V4 hit a 100x inference gain in 44 days — through hardware optimization across GB200 and B200
A Japanese broccoli farmer used Codex to build himself an engineering department from scratch — good "you can just build things" read
Perplexity confirmed a 2028 IPO — won't wait for Anthropic or OpenAI to go first
Standard Bots raised $200M to manufacture robotic arms in the US — as factories race to automate physical work
! OPPORTUNITIES WITH THE TEAMS WE LOVE !
Cursor — Software Engineer, ML Infrastructure
You will work on building and optimizing the GPU infrastructure that powers large-scale AI model training.
Guild.ai — Software Engineer
Work on building and deploying AI agents, focusing on creating and governing large systems.
Fastino AI — Software Engineer
Ship full-stack features end-to-end, owning the full lifecycle from conception to production with minimal oversight. Build and maintain the infrastructure
→ Apply