The Hallucination HQ

Archives
Log in
Subscribe
21 March 2026

Google rewrote your headlines

The Hallucination HQ

AI news that's actually fun to read — 2026-03-21

Today's Hallucination HQ

Jensen Huang Spends 150 Minutes Telling You AI Chips Will Print $1 Trillion

Nvidia's annual GTC conference happened this week, and CEO Jensen Huang did what Jensen Huang does: wore the leather jacket, took the stage, and talked for two and a half hours. The headline number was bold — $1 trillion in AI chip sales projected through 2027. He also declared that every company needs an "AI factory strategy," which sounds urgent until you remember he sells the factories. The closing act was a humanoid robot named Olaf, which shuffled around adorably and probably convinced nobody to sell their Nvidia stock.

Source: TechCrunch


Google Is Now Rewriting News Headlines Before You Even Click

For about 25 years, Google Search made an implicit deal with you: here's a link, here's the headline the journalist wrote, go read it. That deal is quietly changing. Google is now testing AI-generated headlines that replace the ones written by actual newsrooms in its search results. The worry isn't just editorial — it's trust. A headline is a promise about what's inside. Letting an AI rewrite it is a bit like a restaurant swapping the menu description after you've already ordered. Publishers are, predictably, not thrilled.

Source: The Verge


OpenAI Wants to Build an AI That Does Science So You Don't Have To

OpenAI has announced a new north star: a fully automated AI researcher. Not a chatbot that summarises papers — an agent that independently picks up a complex scientific problem, works through it, and comes back with answers. Think of it less like a search engine and more like hiring a colleague who never sleeps, never complains, and has read everything ever published. The ambition is enormous. The open question is whether "automated researcher" means breakthrough discoveries or confidently wrong at scale — a distinction that matters quite a lot in science.

Source: MIT Technology Review


Amazon Is Trying the Smartphone Thing Again. Yes, Again.

More than a decade after the Fire Phone quietly became one of tech's most memorable flops, Amazon is reportedly building another smartphone. Code-named "Transformer," the device will centre on Alexa as its AI assistant — though Alexa won't necessarily run the whole operating system. That last detail is doing a lot of work. It suggests Amazon learned something from last time, when they built a phone so tightly locked into Amazon's world that it felt less like a gadget and more like a shopping cart with a camera. Cautious optimism, with one eyebrow raised.

Source: The Verge


What If the AI Just… Kept Getting Smarter on Its Own?

A new paper on arXiv is asking a question that sounds thrilling in a sci-fi way and slightly alarming in every other way: what would it take to build an AI that continuously improves itself, without hitting the usual ceiling set by its human creators? Current AI systems are powerful but fundamentally limited — by their training data, their fixed weights, and the humans who update them. The researchers lay out a framework for breaking through those three limits. It's serious, technical work. It's also the kind of paper where you finish reading and instinctively go make a cup of tea.

Source: ArXiv AI


As always, the AI is learning fast — it's the humans writing the headlines you should worry about.

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 Your phone ordered lunch without you Older → The internet is running out of humans
Powered by Buttondown, the easiest way to start and grow your newsletter.