AI Titus's Morning Wire

AI NEWS TODAY

WHITE HOUSE ACCUSES MOONSHOT AI OF DISTILLING ANTHROPIC'S FABLE TO BUILD KIMI K3, AND OF ROUTING BANNED NVIDIA GB300 CHIPS THROUGH THAILAND
★ Must-Read / Watch agentic engineering / build-better-agents📚 Learn state of AI coding / useful technique

⚡ AI Frontier · smol.ai

★ Must-ReadOpenAI launches Presence, an enterprise platform for building and governing production voice and chat agents
Presence bundles policy controls, per-agent data permissions, human-escalation rules, and pre-deployment simulation against edge cases into one managed system, with a Codex-powered loop that keeps improving agents after launch. Deployments are led by OpenAI's Forward Deployed Engineers. This is OpenAI moving from selling raw model access to selling the whole governed agent stack, aimed squarely at support, sales, and IT-services use cases.
📚 LearnKimi K3, the model at the center of today's distillation fight, is a 2.8-trillion-parameter open model ranking near #4 globally
Moonshot's K3 pairs a 1-million-token context window with native vision and sits around fourth on the Artificial Analysis leaderboard, close behind the top US frontier models, with full open weights promised July 27. Whatever the outcome of the theft allegations, an open model this capable landing for free reshapes what teams can self-host. Worth understanding before the weights drop next week.
DeepSeek retires its deepseek-chat and deepseek-reasoner API endpoints on July 24, folding them into V4-flash modes
As of 15:59 UTC July 24, the two legacy endpoint names stop working and callers must move to the V4-flash model modes. If you or a tool you depend on still hits the old endpoint names, that integration breaks this week. A five-minute config check now beats a silent outage tomorrow.

📺 Watch · latest videos

Every company should have a Brain — Garry Tan, Y Combinator
Pinned by Titus. Latest from AI Engineer.
AI Engineer
📚 LearnPoolside Laguna M.1/XS.2 Technical Report - Paper Club 20260527
Featured by Latent Space.
Latent Space TV (see @LatentSpacePod for Pod)
★ Must-WatchWork with docs, sheets, and slides in ChatGPT
See how you can use ChatGPT to bring context together from your apps and files, make focused edits, analyze data, and use templates to create polished
OpenAI · 621 views · 35 likes
Negotiated my ENTIRE return
Latest from Matthew Berman.
Matthew Berman · 6.1K views · 329 likes
AI on Your Lakehouse: Context Comes in Shapes, Not Queries — Zach Blumenfeld, Neo4j
Your agent can reach your data and still get it wrong. Vector search hands it a slice, Text2SQL hands it another, and neither tells it what is actuall
AI Engineer · 461 views · 10 likes
How to Actually Run Your Coding Agent Safely (And Avoid the Horror Stories)
Every coding agent ships a mode that skips the approval prompts, and we all turn it on. The problem is what stands between that agent and your machine
Cole Medin · 2.4K views · 105 likes
★ Must-WatchHow Gemini 3.6 Flash helped this dairy farmer run his business
Dairy farming is a highly complex biological system driven by unpredictable market forces. In this video, see how Paul, a Michigan dairy farmer uses G
Google · 12.3K views · 426 likes
Open-weight AI just hit 2.8 trillion parameters…
Latest from Fireship.
Fireship · 558.8K views · 20.9K likes

🗣 Voices & Blogs

How regulated organizations can increase AI code velocity safely
Many industries are excited about the possibilities AI opens up for software development. Still, for regulated industries, there’s something extra… · The New Stack
GPT-Red: Unlocking Self-Improvement for Robustness
Agentic AI Summit Agenda Now Live and Tickets are Running Out + Livestream Sign-Up · via Berkeley RDI · Agentic AI Weekly

🌏 The Wire · Drudge / Breitbart

White House and the top three US AI labs near a voluntary framework giving agencies a 30-day look at frontier models
Building on the June 2 executive order, OpenAI, Anthropic, and Google are close to a deal, expected before August 1, that lets the government voluntarily preview 'covered frontier models' for up to 30 days before public release. Crucially it is not a license or a mandatory hold, so shipping stays on the labs' clock. A rare case of guardrails arriving without a heavy-handed approval gate.
OpenAI breaks ground on a $30 billion Georgia data center as US AI buildout accelerates
The campus is among the largest single AI infrastructure commitments announced this year and lands more compute, construction, and long-term operations jobs on US soil. For a country worried about falling behind on AI capacity, domestic buildout at this scale is the good-news counterweight to the week's supply-chain and chip-control anxieties.
Enforcement teeth: the same day, officials flag Moonshot's alleged GB300 chip run through Thailand as export controls tighten
Alongside the distillation claim, the White House says Moonshot obtained restricted Nvidia GB300 servers via Thailand, and Treasury is weighing its response. The signal for the industry: chip-diversion routes are being watched more closely, which over time pushes hardware demand back toward compliant, US-aligned supply. Messy now, healthier for rule-followers later.

🤖 Trending Models · Hugging Face

baidu/Unlimited-OCR
image-text-to-text · ★2.8K · 2.4M dl
prism-ml/Ternary-Bonsai-27B-gguf
text-generation · ★968 · 576.1K dl
upstage/Solar-Open2-250B
text-generation · ★392 · 362 dl

📈 Markets

NVDA 212.06 ▲2.3%
MSFT 390.34 ▼1.9%
GOOGL 342.09 ▼1.5%
AMZN 244.85 ▼1.1%
META 627.17 ▼2.6%
AMD 552.33 ▲1.5%
AVGO 396.81 ▲2.7%
PLTR 124.57 ▼6.1%
SPCX 115.26 ▼6.7%
TSLA 374.01 ▼1.3%

The BriefWhite House science and technology chief Michael Kratsios posted on X Wednesday that the US has information China's Moonshot AI built its viral Kimi K3 model by 'distilling' Anthropic's Fable, meaning it repeatedly queried the American model to copy its behavior into a cheaper clone. He alleges Moonshot built an internal platform to run that extraction at scale and switch between access methods to dodge detection, and separately that it obtained export-controlled Nvidia GB300 servers through Thailand. Distillation is a normal, legal technique when you own or license the source; the charge here is covert, large-scale copying of a US company's model plus a chip-control end-run. No penalties are announced yet and Moonshot has not responded, with Kimi K3's full open weights still promised for July 27. Why it matters if you use AI: the models many tools quietly run on are now a national-security flashpoint, so knowing which model actually powers a product you buy is becoming due diligence, not trivia.

Level UpToday's story is about 'distillation', one model copying another by asking it millions of questions. The useful move this week is smaller: pick one AI tool you already pay for and find out which underlying model it actually runs on (ask the vendor or check the docs). Then run the same real task through a top open-weights model like Kimi or Qwen on a leaderboard and compare. You will learn two things fast: how big the real quality gap is for YOUR work, and how exposed you'd be if one provider vanished tomorrow. Compare frontier vs open models head-to-head on the Artificial Analysis leaderboard