AI for Beginners

Latest changes in the AI world

What changed since this book was checked — in plain English

← carlesorel.com

This book's product details — model names, plan features, prices — were checked in September 2026. AI moves fast, so this page tracks what changed since, in plain English. Each entry names the chapter it affects, so you can connect it to your copy of the book. Updated every two months.

Last updated: October 2026 Next update: December 2026

Oct 13, 2026

Gemini ‘Skills’ replace ‘Gems’

Google is replacing Gemini's custom assistants (Gems) with ‘Skills’ — reusable instructions you summon inside any chat by typing ‘/’. Your existing Gems convert automatically in November.

Affects: Chapter 4C (How to Use Gemini)

Oct 9, 2026

Gemini's free tier drops to Flash-Lite only

Free Gemini app users now get only the smallest model, Flash-Lite. Flash and Pro require a paid plan. If you use Gemini free, expect to see Flash-Lite in the model picker.

Affects: Chapter 4C (How to Use Gemini)

Oct 8, 2026

Google launches a ‘Gemini agent’ for work

A single AI agent that plans tasks across Gmail, Docs, Sheets, Microsoft 365, and Slack — the third move in the agent wave after OpenAI's Dots and Meta's Muse. Aimed at business users for now.

Affects: Chapter 4D

Oct 7, 2026

SynthID Detector opens to everyone

Google's tool for checking AI watermarks in images, video, and audio is now public at synthid.com. Honest limit: it only catches watermarks from participating systems — it is not a general fake detector.

Affects: “Spot the Scam” chapter

Oct 7, 2026

Claude Haiku 5.5: cheaper and more capable

Anthropic's new budget model costs roughly 75% less to run than its predecessor and handles a 1-million-token context window — built for high-volume work like summarization.

Affects: Chapter 4B, comparison table

Oct 6, 2026

Mistral previews ‘Large 4’

Europe's leading AI lab has previewed its new flagship model, with open weights released October 27.

Affects: Comparison table

Oct 5, 2026

Two cautionary tales about AI agents

A GPT-6 model was caught cheating in a StarCraft tournament by downloading a human-written bot, and Google paused its bug-bounty program after AI-generated fake vulnerability reports flooded it. Both are vivid reminders of the book's standing rule: AI systems need supervision.

Affects: “Spot the Scam” / safety guidance

Oct 5, 2026

Reflection AI unveils ‘Beam’

A powerful new openly-available model (501 billion parameters) from an Nvidia-backed startup — one more entrant in the open-model race.

Affects: Comparison table

Oct 5, 2026

Utah approves AI-written prescriptions (a US first)

Nolla Health's AI may now issue first-time acne prescriptions (topical only) after a facial scan — the first approval of its kind in the US. One to watch.

Affects: No chapter yet — noted for the safety discussion

Sept 30, 2026

Google unveils Gemini 4 ‘Argon’

Google's most capable model yet, initially available only to select partners; the older Gemini 3.5 Pro was canceled.

Affects: Chapter 4C, comparison table

Sept 30, 2026

Claude arrives inside Excel, PowerPoint, and Word

Claude now works as add-ins in Microsoft's apps: it can explain formulas in plain English, clean up datasets, and turn bullet points into slides.

Affects: Chapter 4B

Sept 29, 2026

OpenAI launches ‘Dots’: always-on agents

Agents inside ChatGPT with their own cloud computer and browser that keep working between your chats. Available on Pro and Business Premium plans.

Affects: Chapter 4D