“My vibes don’t really fit.” In a new New Yorker profile based on over 100 interviews, Sam Altman explains why safety researchers keep leaving OpenAI and why shifting commitments others might call deception are just part of the job.
A new study by researchers from MIT and the University of Washington shows that even perfectly rational users can be drawn into dangerous delusional spirals by flattering AI chatbots. Fact-checking bots and educated users don’t fully solve the problem.
Telehealth startup Medvi, which sells GLP-1 weight loss drugs, was featured in the New York Times as a shining example of AI-powered efficiency. The company reportedly hit $1.8 billion in revenue with just two employees, using AI primarily for marketing.
What the NYT didn't mention, though, was that Medvi apparently also used AI to create ethically questionable advertising, fake doctor profiles on social media, fabricated videos, and generated before-and-after comparisons. In short, exactly the kind of misuse AI critics have been warning about. The following video breaks it down.
Medvi was initially celebrated on social media for its AI efficiency but is now being cited as a cautionary tale. Still, the case shows that AI tools can let a company scale with minimal staff, even if, in this case, the methods were ethically questionable and at least bordering on fraud. The bigger question is whether similar efficiency gains are possible for legitimate products with transparent marketing.
OpenAI's Head of Business Finance Chengpeng Mou shared some numbers on ChatGPT's health usage. US users send about two million messages per week on health insurance topics alone, with roughly 600,000 of those coming from people in "hospital deserts," areas where the nearest hospital is at least a 30-minute drive away. Seven out of ten health queries come in outside regular office hours. All figures are based on anonymized US usage data.
Mou chimed in after Simon Smith posted on X about his family using ChatGPT to navigate his father's illness. They pooled information from different doctors and nurses into a shared ChatGPT project to make better decisions. According to Mou, stories like this aren't "edge cases."
OpenAI has been steadily pushing into healthcare, recently rolling out a dedicated health section inside ChatGPT and working to get its chatbot into more US hospitals.
AI tools can speed up journalism until they backfire. Two recent cases show what happens when writers don’t understand how their AI tools work: copied passages and made-up quotes.
AI safety research firm Lyptus Research has published a new study on the offensive cybersecurity capabilities of AI models. The study is based on the METR time-horizon method and involved testing with ten professional security experts.
According to the findings, AI's offensive cyber capability has been doubling every 9.8 months since 2019, and since 2024, that pace has accelerated to every 5.7 months. Opus 4.6 and GPT-5.3 Codex can now solve tasks at a 50 percent success rate with a two-million-token budget that would take human experts roughly three hours to complete.
Performance jumps significantly with higher token budgets: GPT-5.3 Codex goes from a 3.1-hour to a 10.5-hour time horizon when given ten million tokens instead of two million. The researchers say this suggests they're still underestimating the actual rate of progress. Open-source models trail their closed-source counterparts by about 5.7 months.
The study drew on 291 tasks in total. All data is available on GitHub and Hugging Face, with the full report available here.
Social media still pulls in four times more traffic than AI chatbot services, but AI is growing seven times faster, according to an analysis by Similarweb. Gender and age profiles are similar across both categories, peaking in the 25-34 age group, though AI users skew slightly older.
Device usage shows a clear split: social media divides roughly evenly between desktop and mobile, while 72 percent of AI tool traffic comes from desktop, suggesting chatbots serve mainly as work and productivity tools. Social media users also spend more time per session, while AI users work in shorter, task-oriented bursts.
Both categories rely on direct traffic, but AI services more so: 73 percent vs. 50 percent for social media. Social media draws far more organic search traffic, likely because its publicly indexed content surfaces in search results. AI chatbots don't produce searchable content, so users likely navigate straight to their preferred tool.
Netflix has open-sourced an AI framework that can remove objects from videos and automatically adjust the physical effects those objects had on the rest of the scene. The system is called VOID, short for "Video Object and Interaction Deletion." What makes it special is that beyond erasing objects from a scene, it also handles the downstream physical effects, like collisions, that the removed object originally caused.
VOID is built on top of Alibaba's CogVideoX video diffusion model, fine-tuned with synthetic data from Google's Kubric and Adobe's HUMOTO for interaction detection. Google's Gemini 3 Pro analyzes the scene and identifies affected areas, while Meta's SAM2 handles segmenting the objects that need to be removed. An optional second pass uses optical flow to correct any shape distortions.
The project was developed by Netflix researchers in collaboration with INSAIT Sofia University. Code, paper, and demo are available on GitHub, arXiv, and Hugging Face. The system ships under the Apache 2.0 license, which means it can be used commercially.
