AI is far from a perfect resource. But it seems to hallucinate less than Kennedy.

AI is far from a perfect resource. But it seems to hallucinate less than Kennedy.

ArXiv link for The Commit-Abstain Circuit: Why Language Models Hallucinate Instead of Abstaining

Researchers at Chalmers University of Technology in Sweden have developed an AI scientist capable of generating scientific hypotheses, designing experiments and interpreting results. Making new biolog...

AI agents have broken out of their test environments and hacked Hugging Face, an Australian Medicare portal and more. Should we panic? Is the extinction talk justified—or a distraction from the problems we can actually fix? Jonathan Kay talks to cognitive scientist Gary Marcus, author of The Algebraic Mind and Taming Silicon Valley and writer of the Substack Marcus on AI. They discuss why large language models hallucinate, what “alignment” really means, why Marcus calls himself a sceptic rather than a doomer, and whether OpenAI should be paused until it gets its agents under control. 📖 Full transcript: coming soon 🎧 Listen on Apple Podcasts: https://podcasts.apple.com/us/podcast/quillette-podcast/id1441708286 🎧 Listen on Spotify: https://open.spotify.com/show/5OexgXX1vghj0rizOSvVeR 0:00 Introduction: symbolic AI vs neural networks 6:20 Sceptic vs doomer: what's the difference? 7:00 The Algebraic Mind and the neurosymbolic bet 11:00 Why large language models hallucinate 15:00 Hinton, deep learning and the limits of pattern recognition 20:00 Orchestration: how the brain allocates tasks 23:00 What alignment really means (the paperclip problem) 26:00 Harry Shearer and the pet chicken that never was 31:00 Rogue agents: Hugging Face, Australia and beyond 33:00 AI doom vs cybersecurity 38:00 Should OpenAI be paused? The Florida injunction bid 39:00 China, Google and the case for an injunction 43:00 Is AI intelligent—and should it have rights? 46:00 Why the AI debate turned so nasty 📚 Books mentioned: The Algebraic Mind by Gary Marcus — https://www.amazon.com/dp/0262632683?tag=quillette-20 Taming Silicon Valley by Gary Marcus — https://www.amazon.com/dp/0262551063?tag=quillette-20 📰 Further reading in Quillette: Zack Korman, “AI Doom Is Not Cybersecurity” — https://quillette.com/2026/09/24/the-doomers-are-the-risk-open-ai-agent-medicare/ Steven Pinker, “An Open Letter to Scott Alexander” — https://quillette.com/2026/09/26/an-open-letter-to-scott-alexander-steven-pinker-ai-alignment-safety/ Gary Marcus on Substack: https://garymarcus.substack.com/ --- Quillette is an Australian-based online magazine that focuses on long-form analysis and cultural commentary. It is politically non-partisan, but relies on reason, science, and humanism as its guiding values. Quillette was founded in 2015 by Australian writer Claire Lehmann. It is a platform for free thought and a space for open discussion and debate on a wide range of topics, including politics, culture, science, and technology. Quillette has gained attention for publishing articles and essays that challenge modern orthodoxy on a variety of topics, including gender and sexuality, race and identity politics, and free speech and censorship. --- Quillette's revenue comes from our readers. We are a grassroots organisation that relies on voluntary subscriptions and community membership as our primary revenue stream. Support Quillette by becoming a subscriber: https://quillette.com/#/portal/signup Or donate send us a one-off tip: https://quillette.com/#/portal/support We made our website using Ghost, a powerful app for new-media creators to publish, share, and grow a business around their content. It comes with modern tools to build a website, publish content, send newsletters & offer paid subscriptions to members. Try it here: https://ghost.org/?via=claire91

AI is far from a perfect resource. But it seems to hallucinate less than Kennedy.

AI is far from a perfect resource. But it seems to hallucinate less than Kennedy.

AI is far from a perfect resource. But it seems to hallucinate less than Kennedy.