TOGETHER WITH TODAY'S PARTNER
Good morning, {{first_name|there}}. Kimi K3 got so good so fast that Washington is now debating whether Americans should be allowed to run it at all — six days before its weights go free.
Read time: 3 minutes. Same time, every weekday — rate today's issue at the bottom.
🚀 The Big Story: Washington moves to wall off Chinese open AI
Axios reports the administration is actively weighing an executive order and other tools to effectively ban Chinese open-source models like Kimi K3 and Qwen 3.8 inside the US.
Five tools on the table: Entity List designations for Chinese AI labs, federal procurement bans, security advisories, liability rules for companies using Chinese models, and public pressure campaigns.
The enforceability problem: once weights are downloadable, "there is no central switch to flip" — and legal experts flag First Amendment obstacles to any weights-level ban.
The clock: K3's open weights drop July 27; DeepSeek V4 stable lands July 24. Policy is racing distribution, and distribution is winning.
Jason's take: Whatever you think of the politics, the operator math is simple — the cheapest lane of AI just acquired regulatory risk. If part of your stack runs (or will run) on Chinese open weights, you need a fallback plan before July 27, not after an EO drops. Today's Workflow is that plan.
⚡ Quick Hits
A 20B model claims math gold. The new "Loopie" looped-MoE paper (2B active params) claims IMO and IPhO gold without tools — a direct shot at "bigger is always better."
Musk secretly bought a turbine company. An FTC filing shows SpaceX quietly acquired APR Energy (~$1B) — 1GW+ of mobile gas turbines, the same class powering xAI's Colossus. Compute is now vertically integrated with its own power.
Retail is rotating out of the Mag 7 and into AI-adjacent picks like SK Hynix and Marvell, per WSJ — five of the seven now trail the market this year.
The AI buildout has a banker: Morgan Stanley booked $2.3B in H1 capital-markets fees off AI infrastructure deals, up 64% YoY.
Eminent domain, meet data centers. Utilities in Georgia and Pennsylvania are condemning land for AI power lines despite ~70% public opposition — courts are split.
📡 Trending on X
Ben Thompson reset the pricing debate: stop comparing per-token prices — what matters is cost per unit of intelligence, and marginal inference costs are "back in a big way." The thread is everywhere.
WSJ's Israel story is the discourse: a $45M AI influence campaign of fake-sender texts — explicitly designed to reshape how ChatGPT and Claude answer questions — has both AI and politics X arguing about "chatbot SEO."
4,500 Googlers signed a petition demanding layoff protections and hand-delivered it to Pichai — the largest job-security push inside Google yet, and tech X is taking sides.
Vibe-coding demos keep going viral: Ammaar Reshi's prompt-to-deployed-app clips are setting the aesthetic bar for what "building" looks like now.
🛠 The Workflow: Pack your open-weights go-bag
Two big open releases land this week and a possible EO hangs over both. Twenty minutes of prep means you win either way:
Inventory your AI lanes: list every workflow that touches a model, and which model it runs on today.
Pick two fallbacks per lane — one US open-weight option and one API model — so no single ban, outage, or price hike strands you.
Run your top 3 real prompts through both fallbacks today and score the outputs 1–5. Keep the sheet.
Export your assets: prompts, custom instructions, and fine-tune data live in your storage, not only inside one vendor's app.
Calendar July 24 (DeepSeek V4) and July 27 (K3 weights) — re-run the same 3 prompts and let the scores, not the headlines, decide your stack.
Reply with the word "gobag" and I'll send you the scoring sheet template.
🧰 Trending Tools
GLM 5.2 — the open model Hugging Face used to analyze 17,000+ attack events when US models refused. For teams that need an uncensorable analysis workhorse.
Gemini Notebook — NotebookLM's upgrade adds a secure cloud computer that runs code on your sources. For researchers and content teams.
Kimi Code — the coding surface for K3, when signups reopen. For high-volume routine coding on a budget.
Alpha Chat — Current AI's open-source chatbot out of Geneva, backed by $400M in public-infrastructure money. For the privacy-first crowd.
LoRA Speedrun — a public leaderboard for fastest fine-tuning on a single GPU. For tinkerers who learn by racing.
📣 Put your brand here. The AI Innovator reaches AI-first operators, creators, and marketers every weekday. Primary sponsorships are now booking.
That's a wrap
Tomorrow: the countdown to open-weight week — and whether Anthropic answers Kimi K3 before the weights drop.
Enjoying the new format? Share your link — every referral keeps this free:
How was today’s email?
You're reading the 3-minute AI Innovator — same time, every weekday. Hit reply and tell me what you want more of. I read every one.
