🛎️ OpenAI Broke Erdős

Good Morning, AI Enthusiasts!

The ad writes the answer, the travel agent can’t book the trip, the model breaks the theorem, and the “AI vendor” gets fired by coffee.

RECORD

OpenAI Broke Erdős, This Time for Real

👀 What’s happening: An OpenAI internal reasoning model disproved a 1946 conjecture by Paul Erdős, the legendary Hungarian mathematician asking how often n points in a plane can sit exactly one unit apart. Square grids held as the 80-year ceiling. The model built a new construction yielding n^{1+0.014} pairs, a polynomial leap past Erdős's predicted limit. Fields Medalist Tim Gowers told mathematicians on X to sit down before reading.

🌍 How this hits reality: Last October, OpenAI's GPT-5 Erdős claim collapsed when Thomas Bloom, who runs the Erdős database, showed it had just rephrased existing literature. Now Bloom co-signs the verification paper. The fallout runs past math. Every century-old open problem becomes a model run, not a human career. Mathematicians slide from originators to reviewers. The whole economy priced on human intellectual scarcity, from PhD pipelines to vertical AI-for-science startups, just took a haircut.

🛎️ Key takeaway: AI didn't out-think humans on Erdős. It out-lasted them. Every economy built on the scarcity of human genius just lost its scarcity.

RETAIL

Starbucks Fired Its “AI”

👀 What’s happening: Starbucks scrapped its AI inventory tool across 11,000 North American stores, nine months after rollout. The vendor: NomadGo, a roughly 50-person Bellevue startup that in 2024 pivoted from counting people to counting inventory, using phone cameras, LiDAR and on-device CV. Starbucks was its flagship, sold at 99% accuracy and 8x faster than humans. The app kept confusing milk types.

🌍 How this hits reality: NomadGo built what classical computer vision could do in 2017: narrow-trained object detection on a phone. Vision-capable foundation models now read the label on the bottle directly, no per-SKU retraining. NomadGo shipped pre-LLM AI into a post-GPT-4V buying market and never re-tooled.

CEO Brian Niccol's "Back to Starbucks" turnaround bought a vendor that pre-dated this AI era, not one that defines it. Every retailer still riding a 2022-vintage CV vendor contract is reading Starbucks's exit notice. The first AI shakeout is not AI versus humans, it is LLM-native AI lapping pre-LLM AI, with the loser being a company that literally has "AI" in its product name.

🛎️ Key takeaway: The first AI casualties are not human jobs but the AI vendors who slept through the foundation-model upgrade, and the Fortune 500 CEOs who bought "AI" without checking the vintage.

NEW LAUNCH

Airbnb Caught Up with 2023

👀 What’s happening: Airbnb shipped its 2026 Summer Release: in-app car rentals, Instacart groceries, airport pickups, luggage storage, boutique hotels in 20 cities. The centerpiece: an AI assistant that reads a billion reviews, compares wishlist homes, drafts itineraries, and handles support in 11 languages. CEO Brian Chesky called it making "every part of your trip easier."

🌍 How this hits reality: The assistant reads reviews but can't order Instacart. Compares homes but can't book the car. Plans itineraries but can't call the pickup. Airbnb bolted a 2023-shape chatbot to 2026 logistics. Meanwhile Google eats travel discovery upstream: AI Overviews answer trip questions before a click, Maps absorbs the flight-stay-meal layer Airbnb wanted. Chesky assembled every piece of a travel agent, then shipped them as separate buttons next to a chatbot that can only read about them.

🛎️ Key takeaway: The agent era rewards whoever stitches the parts, not whoever owns them. Airbnb owns everything except the agent.

ADS

Google Owns Gemini's Voice

👀 What’s happening: Google is wiring ads directly into Gemini-powered Search. Search "compact espresso pod machine" and Gemini drafts a "custom explainer" for a Nespresso Vertuo Up under "Sponsored Product." Some ads carry an "Ask a question" chatbot pulling answers from the advertiser's site. AI Mode lets sponsored items take over the full reply. Google's VP of ads and commerce Vidhya Srinivasan calls these "next-generation ad formats."

🌍 How this hits reality: The same model that writes the Nespresso "Sponsored" explainer also writes the unsponsored recommendation right beneath it. So Google owes an answer: when you ask Gemini "which espresso machine should I buy?" with no Sponsored tag in sight, what stops the reply from tilting toward whoever pays Google most this quarter? Google can label slots, not tone, ranking instinct, or word choice. The neutrality firewall newspapers and rating agencies built into org charts now lives inside one model's weights, where the firewall is whatever Google says it is.

🛎️ Key takeaway: The fight in AI search is no longer ad placement, it is whether an assistant owned by an ad company can still be trusted to speak straight, and Google just put that doubt under every Gemini answer.

TOGETHER WITH MYCLAW

Personal AI That Starts Personal

A personal AI shouldn’t feel generic. MyClaw lets you install skills that adapt to your preferences from the start, so your assistant feels relevant on day one. Behind the scenes, OpenClaw provides the agent power. What you see is a simpler way to put it to work.

If you’re serious about OpenClaw, don’t run it halfway. Run it the way it was meant to be run — fast, secure, and production-ready by default.

DAILY TL;DR

• Anthropic is reportedly in talks to run Claude on Microsoft-designed AI chips, a fresh sign that frontier labs are looking beyond Nvidia GPUs.
• JPMorgan is rolling out AI tools across its global investment banking business to speed up research, materials, and client work.
• Starbucks scrapped its AI inventory-counting tool across North America after the system repeatedly miscounted and mislabeled store items.
• HHS launched an AI-led audit program to detect fraud, waste, and missing compliance records across federally funded health programs.
• Kawasaki Heavy Industries is partnering with Nvidia on a Silicon Valley robotics center focused on physical AI for medical and mobility robots.
• AdventHealth is deploying ChatGPT for Healthcare to cut administrative work and streamline clinical workflows across its hospital system.
• White House postponed a planned AI executive order after Trump said he did not want new rules to slow U.S. competition with China.

READ MORE

Let the Future Come to Your Inbox

Stay ahead without drowning in information. We turn the most important signals across AI, tech, marketing, and future products into 5-minute reads you can actually finish.

• AI Secret uncovers what really matters in AI
• Bay Area Letters decodes tech and business shifts from Silicon Valley
• Robotics Herald tracks how robots move from labs into daily life
• Marketing Secret breaks down real growth and go-to-market playbooks
• The Hardwire explores hardware, consumer tech, and what’s coming next

TOGETHER WITH US

AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.

We've helped promote over 500 Tech Brands. Will yours be the next?

Email our co-founder Mark directly at mark@aisecret.us if the button fails.