The 12 AI stories that mattered in the week of 19 Oct – 25 Oct 2026
Ranked by impact from 96 stories AIPulsee tracked. Each one explained in plain English.
A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis
A dermatologist compared two leading large language models, GPT-6 Astra and Claude Fable 5.1, to a ten‑year‑old Inception v3 convolutional neural network on 200 biopsy‑proven skin lesions and found the CNN correctly identified the diagnosis for 62% of cases, while the LLMs each succeeded on about 42%. The CNN, fine‑tuned for about twenty minutes on public dermatology images, outperformed the frontier LLMs in recognizing skin cancer from paired clinical and dermoscopic photos.
OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?
OpenAI and Anthropic dramatically reduced the prices of several of their large language models, with OpenAI cutting GPT‑5.6 Luna’s cost by about 80% and lowering Sol’s price by roughly 20%, while Anthropic made the temporary Sonnet 5 discount permanent and introduced cheaper pricing for Fable 5.1 with a lower cache‑read fee. These price cuts bring the cost per token for input, output, and cache reads down significantly across the board.
Cerebras Deploys Trillion-Parameter AI Models
Cerebras explained how it splits trillion‑parameter frontier models, like the 3‑trillion‑parameter Kimi K3, across multiple WSE‑3T wafers, each with 44 GB on‑chip memory, to run inference at 1,000–2,000+ tokens per second. By allocating dense layers with KV cache and linear‑attention layers without it, the system fits the model’s layers onto the wafers and pipelines token flow between them.
AI use leads UC Berkeley math professor to see 'no point' in grading homework
A professor of mathematics at UC Berkeley has expressed that grading homework assignments has become pointless due to the widespread use of artificial intelligence. The headline indicates that the professor feels the effort to evaluate student work is no longer worthwhile because students can use AI tools to produce solutions.
No signs of AI payoff outside the tech sector
Despite massive AI investments and hype, corporate America outside the tech sector has seen no financial payoff: revenue per worker and profit margins remain flat, and there’s no sign of AI boosting productivity. Mainstream firms are treating AI token spending as a cost rather than a growth driver.
America has lost 200k of jobs in media, film and the arts. How much is AI?
America has lost more than 200,000 jobs in media, film and the arts over the past four years, with about 50,000 of those losses occurring in the last year. The decline lines up with the rapid emergence of AI systems that can generate novels, images, music and other digital art at low cost.
Why is AI generated language so hard to understand?
The paper investigates why AI‑generated language feels harder to understand, finding that AI models use more psycholinguistic features linked to processing difficulty—such as inanimate subjects with animate verbs, low‑frequency n‑grams, longer syntactic dependencies, stacked nouns, and reduced object relatives—than human writing or round‑trip translation. The authors propose that this stems from AI models being optimized for information density rather than successful communication, and they demonstrate that applying a length penalty during fine‑tuning pushes models further from human‑like linguistic behavior.
Swedish government offices to ban ChatGPT and Claude
The Swedish government has decided to prohibit the use of two popular AI chat services, ChatGPT and Claude, within its offices. ChatGPT is a conversational AI developed by OpenAI, while Claude is a similar chatbot created by Anthropic.
Every AI model has the same "controversial" take on humans
A study asked 14 AI models for their most controversial opinion about humans, repeating the question three times each. 35 of 41 responses were identical, claiming that humans prioritize belonging or loyalty over truth or principles.
OpenShell – A safe, private runtime for fleets of autonomous AI agents
OpenShell is a runtime that lets fleets of autonomous AI agents operate safely and privately by enforcing strict policies on file access, system calls, and network connections. It uses kernel-level instrumentation to sandbox each agent and only supplies credentials when requests go to approved endpoints.
LLMs are making formal software verification feasible at a much larger scale
Anthropic used its Claude model to formally prove security properties in the compiled RISC‑V code of 123 open‑source libraries, automatically generating assembly‑level proofs that the programs cannot crash or trigger memory‑safety errors under specified conditions. In the process Claude uncovered long‑standing bugs that had escaped detection despite extensive testing.
OpenAI CLI
The official OpenAI command‑line interface (CLI) for the OpenAI REST API is now available on GitHub. It can be installed on macOS via Homebrew, on Debian/Ubuntu/Fedora via a shell script, or by downloading a ZIP for other Linux distributions.