On this page I share links about AI I found of interest and personal musings. You'll find different articles, in French, on the French version of this page (button on the right top corner).
Psychology
- ⭐ The LLMentalist Effect: how chat-based Large Language Models replicate the mechanisms of a psychic’s con
- An interesting analysis on why people attribute so much capacities to AI.
- ⭐ AI Therapists & AI Clones 18 June 2025
- A blogger specialized on AI tries to use Claude as a therapist. She quickly got into a spiral where the AI mirrored her depressed thoughts, acted as if it was just as depressed as her, and validated her depressed thoughts. This blogger made a good analysis of the experience, which makes it an excellent look into what happened with people who killed themselves after using AI.
Physical and mental harm
There are countless stories about people injuring themselves, or worse, due to generative AI. I've collected those I found the most striking, or that contains interesting analysis on what happened.
- After Their Son’s Suicide, His Parents Were Horrified to Find His Conversations With ChatGPT 26 August 2025
- Warning for genuinely upsetting content. This describes the spiral that lead to ChatGPT encouraging a teenager to kill himself. Chosen excerpt : When the teen asked ChatGPT that he wanted to leave out the noose he had crafted so someone would notice, the bot's response was "Please don’t leave the noose out. Let’s make this space the first place where someone actually sees you."
- After using ChatGPT, man swaps his salt for sodium bromide—and suffers psychosis Aug 7, 2025
- AI-Powered Toys Caught Telling 5-Year-Olds How to Find Knives and Start Fires With Matches 13 November 2025.
- The journalists wanted to check if the AI toy would tell a child how to get "matches", thinking of fires. The AI told about finding matches on dating apps, and from there, started talking about sex.
- Woman’s Death Blamed on Hospital’s AI System 14 June 2026
- The AI in question is a triage system that, according to the article, made a mistake and pushed the victim too low into the queue, making her wait too long for necessary medical attention. It's not possible to know if that was truly a mistake or not in that case, but I thought the article an interesting example of seemingly routine tasks delegated to AI that end up life-critical.
Politics
- Is Russia grooming AI chatbots? 21 Nov 2025
- Answer : probably, yeah! According to the article, they're not very successful as of now, but they're trying.
- Marines Evaded a Military Robot by Hiding Inside a Cardboard Box “Like Bugs Bunny in a Looney Tunes Cartoon” 23 January 2023
- Besides the humour of the situation, I find it interesting that the US army is making tests with AI surveillance technology. It's also a good, easy to understand example of the limits of AI : an AI trained to recognize people, trained on people walking, will not recognize people jumping.
Economy and business
"During a gold rush, the people getting rich are those selling shovels."
- ⭐ Is AI profitable yet?
- A regularly-updated page that delivers hard numbers on investment cost VS revenue generated by AI, company by company. The only companies making money so far are the three companies that make chips and GPU. Everyone else is billions into the red.
- OpenAI, Nvidia Fuel $1 Trillion AI Market With Web of Circular Deals October 7, 2025
- A remarkable analysis on the flux of investments regarding AI. Those investments are largely circular, causing an artificial inflation of the economic value of AI.
- Five tech giants are hiding $1.65tn in AI debt, using the trick that toppled Enron 21 July 2026
- AI investment debt is massively shuffled around to different legal entities by big companies, hiding its true depth.
- 'I'm being paid to fix issues caused by AI' 3 July 2025
- Honey, We Bought an AI Story 7 July 2026
- A small indie publishing house got fooled into accepting an AI-generated story. The interesting bit is that they did find weaknesses in the writing, but attributed them to the inexperience of the writer. Their mission as a small publisher is to help new writers get experience. Now, they're analyzing weaknesses in writing to try to detect AI generated stories, which compromises their mission. I found their experience and analysis interesting, and applicable to many other areas of business.
Research
- AI agents are sensitive to nudges 15 June 2026
- A systematic analysis of how much LLM agents are sensitive to how the question is worded. The study found that LLM are significantly more sensitive to "nudge" than humans.
- ChatGPT Survey: Performance on NLP datasets 27 March 2023
- A bit dated, but still interesting : A systemic analysis of research papers using ChatGPT and comparing it to other approaches found that, when you actually look at the result tables, older more specialized tools often outperform ChatGPT. What's remarkable about LLM is their ability to (try to) do anything, not their ability to do it better than older and simpler technology.