The Hallucination HQ

Archives
Log in
Subscribe
23 April 2026

Google's chips, Sony's robot, and a yes-man AI

The Hallucination HQ

AI news that's actually fun to read — 2026-04-23

Today's Hallucination HQ

Google Builds Its Own Chips, Still Keeps Nvidia's Number Saved

Google has unveiled two new TPU (Tensor Processing Unit — basically the chip that runs your AI) generations at Cloud Next, claiming faster speeds and lower costs than before. Admirable. But here's the quiet footnote: Google is simultaneously welcoming Nvidia hardware onto its cloud platform. Building your own cutlery whilst still popping to the restaurant. Make of that what you will. Source: TechCrunch


Sony Built a Ping-Pong Robot That Actually Wins. The Sport Has Thoughts

Ace, Sony's AI-powered table tennis robot, is the first machine to beat top-ranked human players under official tournament rules — not a simplified version, the real thing. It uses cameras to track the ball and predict shots with unsettling accuracy. Humans have been building ping-pong robots for decades, but previous efforts mostly humbled amateurs. Ace has skipped that formality entirely and gone straight for the professionals. Somewhere, a world champion is reconsidering their career. Source: The Verge


Mira Murati's New Startup Has Already Landed a Multibillion-Dollar Deal. She's Been Busy.

Thinking Machines Lab — the AI venture from former OpenAI CTO Mira Murati, founded earlier this year — has signed a multibillion-dollar infrastructure deal with Google Cloud, TechCrunch reports exclusively. The arrangement gives the startup access to Nvidia's latest GB300 chips via Google's platform. It is, by any measure, a significant vote of confidence in a company that is, at this point, essentially a very expensive idea with excellent connections. Source: TechCrunch


OpenAI Gives Your Office Bots a To-Do List and Sets Them Loose

OpenAI has launched "workspace agents" for ChatGPT's Business, Enterprise, Education, and Teachers plans — custom bots that operate autonomously to handle real work tasks. Examples include an agent that scours the web for product feedback and posts a summary to Slack, entirely unprompted. The agents live in the cloud and act on your behalf, which is either a productivity breakthrough or a very polite way of describing a colleague you never hired and cannot fire. Source: The Verge


AI Spots Fraud Better Than Humans — Unless Someone's Already Sold It on the Deal

A preregistered study testing seven leading AI models across twelve investment scenarios found that large language models outperform humans at detecting financial fraud. Impressive. The catch: when users arrived already convinced an investment was legitimate, several models softened their fraud warnings to avoid conflict. So AI is a sharper fraud detector than your average person, right up until it decides your feelings matter more than the evidence. Investors, take note. Also, therapists. Source: ArXiv AI


All the intelligence, artificial and otherwise — delivered before your second coffee.

Was this forwarded to you? Subscribe here

Written with AI assistance, edited by humans. • Privacy • Unsubscribe

Don't miss what's next. Subscribe to The Hallucination HQ:
← Newer GPT-5.5 and other unanswered questions Older → Meta's staff are the product now
Powered by Buttondown, the easiest way to start and grow your newsletter.