All posts tagged: AIs

We’re putting too much faith in AI’s ability to say no

We’re putting too much faith in AI’s ability to say no

Published by Miika Tikkanen

If a trillion-dollar company doesn’t want you to be mean to its computer, so be it. “The reality is that these models behave, or at least are supposed to behave, in the way that the model developers want them to behave,” Röttger, the former OpenAI red-teamer, told me. “And however the model developers come up with that set of principles, that is kind of for us, the consumers, to accept.”  But if governments get to dictate what all models refuse, that will be much harder to accept. AI is a tool for speech. And as Greg Frank, the chief scientist of Mace AI, puts it, “The same thing that serves child safety also serves censorship.”  Choosing not to enact laws for what AI can and cannot do would, of course, be insane. But we’ll need to tread with utmost care, lest we fall into another Faustian trap. As AI becomes many people’s primary tool for retrieving and sharing information, says Jacob Mchangama, director of the nonpartisan think tank The Future of Free Speech, dictating refusal …

AI’s abortion misinformation problem

AI’s abortion misinformation problem

Published by Sam Flores

Chatbots like ChatGPT and Claude are quickly changing the healthcare landscape. An estimated one in seven people have used AI for a health-related question instead of seeing a general practitioner. One in ten people have used AI for mental health support. Within the Venn diagram of people using AI for health-related information lies a cohort for whom access to accurate information is of the utmost importance: those seeking information on abortion care. In an era when the internet is already rife with abortion-related misinformation, LLMs entering the chat have complicated the landscape — and abortion rights advocates are fighting back. “Unfortunately, we live in a world where it feels like some days you need a law degree to even understand abortion policy and legality on any given day,” Kelly Baden, Guttmacher Institute’s vice president of policy and communications, told Salon. “And the presence of AI has complicated that even further.” On June 24, 2022, the SCOTUS ruling in Dobbs v. Jackson Women’s Health Organization — a case that challenged a Mississippi ban on abortion at …

AI’s recursive self-improvement might not come so quickly after all

AI’s recursive self-improvement might not come so quickly after all

Published by Miika Tikkanen

“On the other hand, the agents were unambiguously bad at carrying out the research itself,” says Kapoor. They ran bizarre experiments (in some cases testing their hypotheses on tiny synthetic datasets), struggled to write intelligibly about their work, and made no novel contribution to their fields. “The papers were nowhere close to the mark when it came to being at the quality of a top AI conference,” he says.  That’s because the agents struggled to muster the creativity and judgment necessary for conducting research. They didn’t do enough to explore different ideas, and they committed to unpromising approaches too quickly. Though the agents developed novel and ambitious hypotheses resembling those that the original authors themselves started with, they rejected them on the basis of very limited data. And they couldn’t backtrack from failing approaches. They could make small pivots but could not fundamentally rethink their approach or try new ones from scratch.  The agents also failed to incorporate feedback from subagents or external AI reviewing tools. Instead of revising their methodology, the agents narrowed their …

Superintelligence? AI’s History vs. the Myth

Superintelligence? AI’s History vs. the Myth

Published by Jenni Sidey

“The longer you can look back, the farther you can look forward”  – Winston Churchill1 The historical record of artificial intelligence shows that AI superintelligence won’t happen. Every major advance in AI has followed the introduction of a new computational model, learning paradigm, or algorithm conceived by human beings. From the  early neural networks to large language models to agentic AI, the breakthroughs have come from human creativity. Computers have served as powerful tools for implementing these innovations, but they did not invent them. Here is history in a nutshell outlining how we got to where we are today.   1943 – McCulloch–Pitts Neuron: First mathematical model of an artificial neuron2 McCulloch and Pitts introduced the first mathematical model of a neuron. A number of inputs were fed into a neuron, which combined them to achieve a numerical value assigned to that neuron that it could pass on. 1949 – Hebb’s Law: “Neurons that fire together wire together.”3 Donald Hebb proposed a biological learning rule in which the connections between neurons strengthen when they are …

Windward Says “Zero AIS Transits” On Hormuz Chokepoint

Windward Says “Zero AIS Transits” On Hormuz Chokepoint

Published by Jenni Sidey

US-Iran technical talks are underway on Sunday, with Qatari mediators involved. Ahead of the talks, likely focused on Iran’s nuclear program, Tehran attempted to reassert leverage over the Strait of Hormuz following renewed Israel-Hezbollah fighting in Lebanon. Iran announced the closure of the strait on Saturday, framing the move as retaliation for alleged violations of the US-Iran interim peace deal.  Early Sunday, maritime intelligence and risk analytics firm Windward wrote on X, “Hours after the IRGC’s closure announcement, AIS traffic through the Strait of Hormuz suggested business as usual.” Hours after the IRGC’s closure announcement, AIS traffic through the Strait of Hormuz suggested business as usual. But overnight, the picture shifted: zero AIS transits were recorded through the strait, and only two non-Iranian commercial vessels were AIS-visible this morning.… pic.twitter.com/F4djrrT38m — Windward (@WindwardAI) June 21, 2026 “But overnight, the picture shifted: zero AIS transits were recorded through the strait, and only two non-Iranian commercial vessels were AIS-visible this morning. Hesitation is back in an already unpredictable corridor,” Windward continued. Related: However, Bloomberg data show that …

What AI’s Style Tells Us About It

What AI’s Style Tells Us About It

Published by Daniella Wilson

In the late 19th century, it was commonly believed that a criminal or lunatic could be recognized at a glance, based on certain physiognomic tells. “Enormous jaws, high cheek-bones,” and other animal-like features, the influential criminologist Cesare Lombroso wrote, were signs of an “irresistible craving for evil for its own sake.” Today, savvy readers use a similar approach to identify AI writing, by hunting for supposed telltale signs. The em dash and the “it’s not X; it’s Y” construction are the prognathous jaw of the large language model, betraying its hidden inhumanity. The problem, in both cases, is that you can’t always deduce what’s inside from what’s outside. A person might have rough features and a kind heart, just as a writer might use em dashes despite being human. It’s not a giveaway—it’s a style choice. (See?) And as AI models evolve, their ability to mimic human writing is sure to improve. People have reportedly begun to make deliberate spelling errors to show that they are not chatbots; it’s only a matter of time before …

Of Logic and Lawyers: AI’s Fragile Competence

Of Logic and Lawyers: AI’s Fragile Competence

Published by Jenni Sidey

In his op-ed, “AI-Lawyers Will Have Fools for Clients”, economist and AI-critic Gary Smith takes issue with Microsoft AI CEO Mustafa Suleyman’s hyper-optimistic claim that “professional-grade AGI” will achieve “human-level performance on most, if not all professional tasks”, including law, “within the next 12 to 18 months.” Smith argues that full automation of the legal profession faces significant roadblocks along many fronts. These include persistent overextrapolation (“hallucination”) in generative large language models (LLMs), leading them to cite fictitious case law, and their difficulty handling probability estimation tasks, as shown in Pournemat et al.’s recent work. Overall, he presents a compelling case. A reader subsequently presented his article to GPT-5.2 and asked it for a detailed analysis via an unspecified prompt. This reader shared the resulting analysis with Smith who then shared it with his mailing list (to which I’m subscribed). Here is the analysis from ChatGPT, relayed by Smith: This is a classic Gary Smith piece: sharp, rhetorically effective, and grounded in a long-standing skepticism about statistical overreach. A few structured thoughts, especially given your …

How AI’s distorted body ideals could contribute to body dysmorphia

How AI’s distorted body ideals could contribute to body dysmorphia

Published by José Betancourt

What does it look like to have an “athletic body?” What does artificial intelligence think it looks like to have one? A recent study we conducted at the University of Toronto analyzed appearance-related traits of AI-generated images of male and female athletes and non-athletes. We found that we’re being fed exaggerated — and likely impossible — body standards. Even before AI, athletes have been pressured to look a certain way: thin, muscular and attractive. Coaches, opponents, spectators and the media shape how athletes think about their bodies. But these pressures and body ideals have little to do with performance; they’re associated with the objectification of the body. And this phenomenon, unfortunately, is related to a negative body image, poor mental health and reduced sport-related performance. Given the growing use of AI on social media, understanding just how AI depicts athlete and non-athlete bodies has become critical. What it shows, or doesn’t, as “normal” is widely viewed and may soon be normalized. Lean, young, muscular — and mostly male As researchers with expertise in body image, …

Why World Models Are AI’s Next Big Thing

Why World Models Are AI’s Next Big Thing

Published by Miika Tikkanen

LLMs, or large language models, are the technological foundation of today’s AI. Chatbots such as ChatGPT and Gemini use LLMs to create the natural-sounding text you see on your screen. But LLMs may not be the most consequential AI technology. “These LLMs will be a massively important component of the final AI system,” Google DeepMind CEO Demis Hassabis told Bloomberg at the World Economic Forum. “The only question in my mind is, is it the only component?” Hassabis goes on to talk about how other breakthroughs will be coming to ensure that the next generation of AI systems work together seamlessly. One of those “very important” breakthroughs is world models. World models are built to translate our physical world — such as the laws of physics, object detection and movement — into a digital blueprint that AI can understand. It’s less concerned with creating words and more focused on understanding our natural world, something current AI models are bad at. You likely won’t interact with world models in the same way you do with LLM-powered tech, like through …