The Hallucination HQ

Archives
Log in
Subscribe
23 August 2026

Harvard's $699 AI professor, reviewed

The Hallucination HQ

AI news that's actually fun to read — 2026-08-23

Today's Hallucination HQ

The New Kid Outscored the Prefects

British AI lab Inherent — founded by DeepMind alumni who apparently had unfinished business — has released Faraday, an AI agent built to replicate scientific research papers. In benchmarks, it outperformed models from both Anthropic and OpenAI on this task, which is either impressive or a reminder that incumbents get comfortable. Replicating existing research isn't glamorous, but science that can't be reproduced is just expensive fiction. A solid first act. Source: TechCrunch


AI Conquered Chess. The Factory Has Other Ideas.

Automation Taipei 2026 laid bare an awkward truth: AI can write poetry, pass bar exams, and beat grandmasters — yet getting a robot to reliably pick up an oddly shaped widget in a noisy factory remains genuinely hard. The physical world, it turns out, did not get the memo about the intelligence explosion. Varying lighting, inconsistent parts, and the sheer stubborn chaos of manufacturing mean the factory floor is humbling the same technology that makes everyone nervous at dinner parties. Source: Digitimes


Harvard Will See You Now. Well, Sort Of.

For $699, Harvard Business School's new Foundry programme offers aspiring founders startup coaching — including AI avatars of its instructors ready to critique your pitch at any hour. It's a sensible idea, genuinely. The avatar won't be distracted, won't check its phone, and critically, won't bill by the hour. Whether a digital facsimile of a Harvard professor carries the same psychological weight as the real one is a question worth $699 to find out, or considerably more not to. Source: TechCrunch


It's Not Stupid. You're Just Not Feeding It Properly.

A well-argued post doing the rounds explains why locally-run language models — AI that runs on your own machine rather than the cloud — often seem underwhelming. The culprit is usually configuration, not capability: wrong settings, inadequate memory allocation, poorly chosen model sizes. The model isn't dim; it's running a marathon in Wellington boots. A few adjustments, the post argues, can transform the experience considerably. Turns out "my AI is rubbish" is sometimes "I didn't read the instructions." Source: Hacker News


Your Next App Has No Buttons. Condolences to Buttons.

The argument that voice is becoming the dominant way we interact with AI is gaining traction — and momentum. You can now talk to your documents, your calendar, your photos. The keyboard, once thought indispensable, is quietly being made to feel unwelcome. The piece raises privacy concerns worth taking seriously: a device that's always listening is a commitment most people are making without quite realising they've made it. Convenient, yes. Unprecedented, absolutely. The microphone has entered the chat. Audibly. Source: Livemint


As ever, the machines are learning. Whether we are is a separate question.

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 The AI that won't say who made it Older → Claude got a little too friendly
Powered by Buttondown, the easiest way to start and grow your newsletter.