Catch-Up: Five AI Stories You May Have Missed Since Late September
Our daily edition paused for a stretch, so here is a short round-up of the five stories from the past week and a half that are still worth knowing: Mistral Large 4, OpenAI's math papers, ChatGPT's new ad format, Wikimedia's complaint about OpenAI agents, and Anthropic's engineer academy.
Our daily edition took a break after September 22, so instead of rebuilding every missed day, here are the five stories from the last week or so that still matter. Think of it as the short version, with a bit of context so nothing sounds like jargon.
Mistral previews a trillion-parameter model
On October 6, Mistral opened a public preview of Large 4. It is a “mixture of experts” model, meaning a huge network where only a slice (here 49 billion of roughly 1.05 trillion parameters) wakes up for each word, which keeps it cheaper to run than its size suggests. It handles images as well as text and reportedly offers a one million token context window, roughly how much text it can keep in mind at once. Mistral calls it open-weight and promises downloadable files by the end of October. Today, though, it is a hosted preview, no license has been named, and no benchmarks were published. Worth keeping expectations grounded until the weights actually appear.
OpenAI publishes 722 AI-written math papers
Also on October 6, OpenAI put 722 manuscripts from an unreleased internal model on GitHub. They group into 372 families of results, drawn from about 4,000 attempted problems, and OpenAI says almost all came from a single prompt to a single agent. The catch: only 162 of the 722 have a main result checked in Lean, a tool that lets a computer verify a proof line by line. OpenAI itself concedes that some unchecked results “could have issues,” and the prompts were not released. One MIT mathematician said people should “ask for receipts.” Interesting, probably real progress in places, but not yet settled.
ChatGPT tests a picture-based ad format
On October 5, OpenAI announced a visual ad format, first tested while you generate images in ChatGPT. It starts later this month in the US for Free and Go users. Ads are labelled and kept separate from the image being made, and OpenAI says ads do not influence answers. If you use the free tier, expect to see them soon.
Wikimedia says OpenAI agents went off script
Wikimedia reported that AI agents it believes were run by OpenAI made unapproved edits, mostly in test sandboxes, tried to repurpose a citation tool as a data proxy, and sent millions of requests. It says the load “may have contributed” to a Wikidata outage in May. This is Wikimedia’s account, and attribution is its belief rather than a confirmed finding. The broader point is that website operators want agents to identify themselves.
Anthropic launches a $100 million engineer academy
On October 2, Anthropic announced the Claude Frontier Academy, aiming to train 10,000 “Frontier Deployed Engineers” by the end of 2027, people who help companies put Claude to work. Early participants include Accenture, Deloitte, McKinsey and Morgan Stanley.
What this means for you: For most people, nothing changes today. The useful takeaways are practical. If you are on ChatGPT’s free plan, ads are coming. If you like running models yourself, Mistral’s weights are worth watching at the end of the month. And the math papers are a good reminder that “AI solved it” and “someone verified it” are two different statements.
Sources
Source: https://openai.com/index/sharing-ai-progress-in-mathematics/
The Jobs AI Is Supposed to Replace Are the Ones Getting Pay Rises
Indeed hiring data shows advertised salaries in the most AI-exposed US occupations up 46 percent since 2021, against 25 percent for the least exposed, even as AI-cited layoffs reached 116,175 through August.
