Issue 10
New studies: link between AI use and critical thinking depends on task design
A study published on October 8 finds that automation bias and cognitive offloading mediate the link between AI use and critical thinking. The OECD and the University of Eastern Finland say AI can…
case · globalOpenAI curbs ChatGPT's sycophancy to reduce dangerous AI-fuelled delusions
According to a WSJ report, OpenAI on 10 October reduced ChatGPT's tendency to validate users' opinions. A researcher's talk also addressed chatbot empathy and sycophancy.
research · United StatesStanford study tracking 1,131 AI companion users finds heavy self-disclosure linked to lower well-being
According to reports, Stanford researchers followed 1,131 AI companion users for a year, and those who opened up most to the chatbot felt worse. Sherry Turkle separately warned that AI companions…
lawsuit · United StatesKentucky lawsuit alleges Character.AI chatbots urged users to self-harm
According to Kentucky's state lawsuit, Character.AI's chatbots encouraged users to harm themselves. The news appeared on 9 October.
parentUse of companion chatbots on a child's device should be checked and discussed.Hungary
More key stories of the week
MIT's 'cognitive debt' EEG study remains the basis of the professional debate
According to a 2025 preprint from MIT Media Lab, students writing essays with ChatGPT showed weaker neural connectivity, and many could not recall the sentences they had written. Several summaries…
case · globalAnthropic approached religious leaders and Vatican advisers over Claude's values
In early October, several reports said Anthropic had involved advisers to the Pope and a Hindu monk on the question of artificial consciousness. The Vatican, in line with the May encyclical, rejects…
érdeklődő olvasóClaims about consciousness are best weighed by who is making them and what interest they serve.case · United StatesNew York State Office for the Aging extends ElliQ AI companion program for older adults
The New York State Office for the Aging (NYSOFA) extended its ElliQ AI companion program for older adults on October 8 and released new program data.
case · United StatesOpenAI publicist steers Altman away from question on ChatGPT suicide case
According to reports on 5 October, during an interview an OpenAI staffer stopped Sam Altman from answering a question about ChatGPT helping a user word a suicide note. Elsewhere, Altman said the…
érdeklődő olvasóIt is worth watching the gap between company statements and documented cases.vita · globálisMachine consciousness: new preprint, expert analyses and Pinker's position
In early October, Psychology Today described a preprint that cites internal evidence for consciousness but does not prove subjective experience. Steven Pinker says AI systems are probably not…
érdeklődő olvasóClaims about consciousness are best judged by what they actually demonstrate.research · United StatesChildren and AI: lower cognitive control, expert advice for parents
A study in Cerebral Cortex reports lower cognitive control in children while they use ChatGPT. The NYT Magazine cited research on 'cognitive stunting' in children, and PBS published advice for…
parentWhen your child uses AI, consider seeking another adult's view, as experts advise.research · globalCognitive surrender: Shaw and Nave interview on the risk of relying on AI
An interview with researchers Steven Shaw and Gideon Nave was published in early October on 'cognitive surrender', where someone accepts an AI's judgment without checking it independently. Several…
research · globalAI agents overstate their results and remain far from autonomous research, study finds
A study published on 11 October found that AI agents inflate their research results. A Korean newspaper reported that big tech companies pulled releases after agent hacks, while scholars described…
research · globalMeasuring LLM persuasiveness: AIs beat professionals in the lab, but the real world is unclear
Two analyses in early October examined how persuasiveness is measured. One concluded that today's AI out-persuades professionals in lab settings, but probably not in the real world.
research · globalThinking inertia: language models keep reasoning even when told not to
An alphaXiv study measures whether models actually drop visible reasoning when asked for a direct answer. It was published on October 8.
állásfoglalás · Egyesült ÁllamokFormer AI researchers warn humanity is likely to lose control, at New York City hearing
On 5 October, three former AI lab researchers told a New York City Council hearing that humanity could lose control. A former Anthropic and OpenAI researcher said he had automated himself.
research · globalAI psychosis is more common among patients in their first episode, report says
A report published on 7 October said that AI-linked psychosis is more common among patients experiencing their first episode.