LATEST DEVELOPMENTS
GEMINI
Gemini 4 Argon Really Surprised Me
Image: GPT Image 2.5 / PANONDA
The fact — Google’s Gemini 4 Argon became generally available on October 13, 2026, after restricted access beginning September 30.
The details — The API costs $2 per million input tokens and $10 per million output tokens through November 30, rising to $4 and $20; its context window is 1,048,576 tokens.
Why it matters — Publishers can feed Argon text, images, audio, video or PDFs and use it for long-form drafting, revisions, research and small app creation, but API costs double on December 1.
Source: Two Minute Papers
QUICK HITS
- Meta shakes up the frontiers, open and closed — Interconnects (Nathan Lambert)
- [AINews] High Taste Testers Ep. 1 — Latent Space
- Quoting Addy Osmani, Morgan Westlee Lunt, and Qing Wang — Simon Willison
- DeepSeek’s new Flash ⚡, Replit drops Anthropic 🔄, Anthropic’s workforce 💼 — TLDR AI
- Inline rendering with document.currentScript — Simon Willison
THE MAIN STORY
Is it legal to train AI on your work? An appeals court answered for the first time — with an asterisk
The fact — The Third Circuit affirmed that ROSS Intelligence can't claim fair use for training on Westlaw headnotes. It's the first US appellate ruling on AI training and fair use.
The details — On Tuesday, Sept. 29, a three-judge panel of the Third Circuit affirmed that ROSS Intelligence — a now-defunct legal research startup — infringed Thomson Reuters' Westlaw headnotes and could not claim fair use for using material derived from them to train a competing AI search tool.
Why it matters — It also explicitly isn't about generative AI. read it
TESTED
Checks out, with a caveat · Free ChatGPT is "unlimited" only for everyday text — image, voice and upload have a cap
Reels say the Free plan became unlimited. OpenAI's official page says "unlimited everyday text chats" — and lists separate limits for image generation, advanced voice, uploads and search.
IN THE LOOP
- Bilawal Sidhu: This 4D tech feels like a Time Machine. I call it the Geospatial Memory Palace.
- The Rundown AI: SoftBank’s $18B bet on robot guards
- Matt Wolfe (YouTube): 3 New Claude Prompts You Need To Try
FREE THIS WEEK
Practical Deep Learning for Coders — fast.ai
Train and deploy state-of-the-art deep learning models, code first. 9 lessons, ~90 min each, needs about a year of coding.