The Hallucination HQ
AI news that's actually fun to read — 2026-05-18
|
Today's Hallucination HQApple's Privacy Play: Siri Will Forget You Ever Asked That
Apple's forthcoming chatbot-era Siri, arriving in iOS 27, will let users auto-delete conversation history — a feature designed to lean hard into Apple's privacy reputation while the company quietly catches up to rivals who've had a two-year head start. It's a clever move: when you can't win on intelligence, win on discretion. Like a butler who's slightly slow but absolutely will not gossip. Whether privacy alone shifts the AI loyalty of users already deep in the Google or OpenAI ecosystem remains the actual question.
Source: The Verge
AI Won't Speed Up Your Broken Process — It'll Just Break It Faster
A widely-shared blog post is making the rounds with a thesis that will annoy many a consultant: AI doesn't accelerate processes, it amplifies whatever is already there. Feed it a well-designed workflow and you get efficiency. Feed it a bureaucratic nightmare and you get a faster bureaucratic nightmare, delivered confidently and at scale. The post has 574 upvotes on Hacker News, suggesting that people who build software have quietly known this for a while and are relieved someone finally wrote it down.
Source: Hacker News
The Bias Is Still In There, It's Just Being Polite About It
New research has found that instruction-tuned language models — the ones trained to behave nicely — can produce superficially fair outputs while their internal representations remain quietly biased. Think of it as a hiring manager who's learned exactly what not to say in an interview. The study specifically examines whether these suppressed biases can still influence real decisions in high-stakes contexts, like hiring or lending. The answer, in the cautious language of academic papers, is: yes, and we should probably look into that.
Source: ArXiv
Eric Schmidt Discovers That Commencement Crowds Have Opinions
Former Google CEO Eric Schmidt delivered the commencement address at the University of Arizona on Friday and was booed — repeatedly — whenever his remarks turned to AI. This is perhaps not shocking when your audience is a cohort of graduates entering a job market that has spent the past two years being described, by people like Eric Schmidt, as something AI will transform. He kept going, which takes either courage or a very good noise-cancelling internal monologue. The students, for their part, have presumably been rehearsing for a difficult labour market.
Source: The Verge
An AI That Predicts Outbreaks, Without Requiring an Expert in the Room
Researchers have developed an LLM-guided system that can autonomously forecast the spread of infectious diseases across multiple pathogens — without requiring expert epidemiologists to manually build each model from scratch. The system uses a tree-search approach, meaning it explores and evaluates different forecasting strategies on its own. Given that the traditional method bottlenecks badly when trying to scale to new regions or emerging threats, automating even part of that process is genuinely useful. Public health, it turns out, is one domain where faster and less labour-intensive are both welcome at the same time.
Source: ArXiv
|
|
All the intelligence, none of the warranty.
|
|
Was this forwarded to you?
Subscribe here
Written with AI assistance, edited by humans. •
Privacy •
Unsubscribe
|
|
