On-device models just passed GPT-4 ยท M&A Beginners ๐
| View this email in your browser |
![]() Models & Agents for BeginnersAI explained simply โ for beginners and teens.
|
By the numbers
|
๐ง If you only have 10 minutes this week Episode 200 ยท AI just turned classic short stories into playable 3D worlds you can walk through right now. 2026-10-18 โถ Listen now |
This Week's Big PictureA year ago, GPT-4 was the model everyone treated as unbeatable, and using AI mostly meant typing into a chat box. This week that picture cracked in several places at once. Nineteen models small enough to run on a regular smartphone scored higher than GPT-4 on standard tests. ChatGPT stopped being only a conversation partner and started acting: it drafted the messages that got Adobe to waive an early-termination fee of about AUD 359, and a new Chrome extension lets it read every open tab and fill out forms across them. Voice mode in six languages -- French, German, Hindi, Mexican Spanish, Korean, and Japanese -- finally sounds like a person instead of a robot that keeps pausing to check. The other half of the week was about making things you can actually step into. A system called Argon took one short paragraph -- a vast, unknowable presence arriving in a quiet suburb, not horror, not sad -- and built a playable browser game. Google's Argon model also produced a five-stage interactive biology atlas that lets you scroll from a sunlit leaf down into the chloroplasts that capture sunlight. Researcher Ethan Mollick asked Fable 5.1 to turn Jorge Luis Borges short stories into walkable 3D worlds. Game design, science class, and literature all got the same upgrade: a clear idea can become something you move through, not just read. Put those threads together and the narrative is clear. AI is leaving the chat window and sitting next to you -- on a phone with no signal, in the browser with all your tabs open, in customer-service chats, in the voices you use to practice a language. That is exciting for homework, hobbies, and first jobs. It is also why privacy settings, on-device models, and websites that block AI helpers showed up in the same seven days. We closed the week on episode 200, a neat milestone for a show that still talks to beginners. The skill to practice now is not only how to prompt, but when to let this teammate act, and what you keep off-screen. Top Stories1. On-device models overtake GPT-4 Nineteen AI models can now run on a regular smartphone and beat GPT-4 on the tests that made it famous. GPT-4 lived on OpenAI's servers and held the top spot for a full year; these models live in your pocket and still win. Ethan Mollick flagged the shift, and you can already try free or low-cost on-device AI or local LLM apps on Android and iOS, often with no internet at all. โถ Episode 195 ยท 2026-10-13 Why it matters: Essay feedback, coding hints, and story help no longer depend on a signal or a paid plan, and learning to use local models will be a useful skill sooner than most people expect. 2. ChatGPT can see your tabs and act on them OpenAI updated its Chrome extension so ChatGPT can read every open tab, summarize the pages, compare products, and fill out forms using Computer Use in Work mode. The upgrade is available on every ChatGPT plan, free and paid. It is like a super-organized friend who can glance at all your windows at once -- and type into them. โถ Episode 199 ยท 2026-10-17 Why it matters: School research and shopping get faster, but anything on screen is now in play, so close sensitive tabs before you turn the helper on. 3. A chatbot talked Adobe into dropping a cancellation fee A Reddit user asked ChatGPT for help canceling Adobe Creative Cloud, then copied its polite, specific messages into the support chat one by one. The model asked for the exact end date, any fees, and a goodwill waiver; Adobe waived about AUD 359, confirmed zero licenses remaining, and left access running until 31 October 2026. โถ Episode 194 ยท 2026-10-12 Why it matters: Everyday people are already using AI as a patient teammate for real money-and-paperwork tasks -- a preview of computer use outside the browser. 4. One paragraph becomes a playable game -- and a living textbook Argon took a prompt about Imminence -- something huge and strange arriving in a suburb, without horror or sadness -- and assembled a full playable game in the browser. The same approach powers a free five-stage biology atlas that starts at a sunlit leaf and zooms into chloroplasts, so photosynthesis becomes something you can pause and explore. โถ Episode 196 ยท 2026-10-14 ยท โถ Episode 197 ยท 2026-10-15 Why it matters: Making a game or explaining a science chapter used to mean teams, months, and special software. A clear description can now become something you click through on a school laptop. 5. Borges short stories you can walk through Ethan Mollick asked Fable 5.1 to turn Jorge Luis Borges short stories into interactive 3D scenes that capture the feeling of each piece, not just illustrate it. The worlds are free to explore in a browser on a phone or laptop, no account required. โถ Episode 200 ยท 2026-10-18 Why it matters: For English, art, or creative writing, this is a glimpse of books and history lessons that behave more like small adventures than pages of text. Trend WatchFrom answers to actions. ChatGPT writing support-chat scripts, filling forms across tabs, and speaking in new voices are the same shift: models that do a task in the world, not just describe it. Computer use is the phrase to learn this month, because it is how chatbots become coworkers. Creation collapsing to a prompt. Argon games, the biology atlas, Fable's 3D Borges worlds, and a teen who built a full game with AI helpers all say the same thing. The bottleneck is no longer game engines and modeling skill -- it is taste, and a clear description of the feeling you want. Power moving onto the device -- and meeting resistance. Local models that beat GPT-4 keep your prompts off distant servers. At the same time, some websites are starting to block AI helpers, and Anthropic is letting third-party experts look deeply at Claude's safety. Capability, privacy, and pushback are arriving together. Quick Hits
What to Watch Next WeekWatch whether Computer Use stays a Chrome novelty or spreads into everyday school and shopping workflows -- and how quickly people learn to close the sensitive tabs first. Watch the app stores for those on-device models; if they are truly GPT-4-class in your pocket, homework and travel without data will feel different fast. And watch the cat-and-mouse with websites blocking agents: if Fable, Argon, and ChatGPT keep acting on the open web, more sites will decide whether helpers are welcome. Safety access at Anthropic is also worth tracking: as agents get more power to act, outside review will matter more, not less. If this roundup helped, send it to someone who still thinks AI is only a chat box, and listen to the full episodes at nerranetwork.com. |
|
๐ฌ Reply to this email โ Patrick reads every one. Share: X ยท LinkedIn ยท WhatsApp Forwarded this email? Subscribe here โ it's free. |
๐บ Watch on YouTube ย ยทย ๐ Read the blog ย ยทย ๐ผ Free image gallery (CC BY-SA) ย ยทย ๐ Data Hub & Story Trackers ย ยทย ๐งญ Start Here Nerra Network ยท AI-narrated voice (Grok TTS) ยท Editorial by Patrick You're receiving this because you subscribed to Models & Agents for Beginners on nerranetwork.com. |
