Together with Claude I made a very large web app where a student can train a language model and apply a variety of mechanistic interpretability tools. The app includes an AI guide that reacts to events, gives tours, and answers questions surprisingly well. docs.google.com/document/d/1...

I noticed a discrepancy in the Nature article “Generative artificial intelligence creates delicious, sustainable, and nutritious burgers” and discussed it with Claude. It decided on its own to find and run the code to get to the bottom of the matter. docs.google.com/document/d/1...

With Claude’s help I added AI guides to three more apps, one of which I’ve never used before. In the process Claude gained enough experience to generate resources to improve the process of adding guides. docs.google.com/document/d/1...

I asked Claude to add an AI guide to a 2700-line app. As the enhanced app is run the guide explains and highlights sections. The user can ask questions and steer to the demo. docs.google.com/document/d/1...

Claude helped me build an app that given two Logo turtle programs creates a program that can interpolate (and extrapolate) between them. In 2D or 3D. docs.google.com/document/d/1...

When I saw 2 butterflies dance and then fly apart, I had questions. Thinking of Seymour Papert’s 30-year old vision of a knowledge machine, I instigated a conversation with Claude. docs.google.com/document/d/1...

Rather than search through 60 Google Docs that needed updates to some links I asked Claude Cowork to do it for me. A short prompt was all I needed. docs.google.com/document/d/1...

Claude and I designed a game which views language as a microworld, a place where learners grow a construction kit. Communication is facilitated by defining words in terms of the few words the aliens know. There is a purely verbal version. docs.google.com/document/d/1...

With Claude’s help I made a customized version of Claude. docs.google.com/document/d/1... A test I ran was comparing the process of co-creating an app with the personalized and default versions of Claude. The apps they created fuse together different LLM responses. docs.google.com/document/d/1...

Everyone should tinker with AI. I explored asking ChatGPT to act like a journalist who is interviewing me. Perhaps, if I persisted, it would have resulted in a decent article but it didn’t work very well. But it was worth trying. docs.google.com/document/d/1...

I made an app for children to design and experiment with agentic systems that learn in a constructionist manner. Students can construct agents that create images, poetry, and short stories. Claude did much of the design and all of the implementation. docs.google.com/document/d/1...

Twenty years ago I made a “metagame” for students to construct space games like Lunar Lander. With the help of both Codex and Claude Code I reconstructed the program. I then created a fresh design and implementation with ChatGPT. docs.google.com/document/d/1...

Computers can use reinforcement learning (RL) to learn how to play games. I asked Claude to create an app for children where they can explore RL on Tic-tac-toe, Connect Four, and Pong. I’m impressed how well this turned out. docs.google.com/document/d/1...

Claude and I made a kid-friendly app for kids to explore the evolution of little Logo programs. It supports both artificial and natural selection. It includes an ask-the-AI feature that works very well. docs.google.com/document/d/1...

I asked Claude to make a modern version of the 1979 TT2500 variant of Logo by adding color, pen width, and more. I programmed an animated aquarium in it. I worked with ChatGPT to create a 2200-line Python program to add music to my 1979 MANDALA program docs.google.com/document/d/1...

You CAN code. Even young children can have a conversation with a chatbot and end up with a web app that does what they want. I’ve been posting about this for 3 years and wrote a book on the topic. WIRED used to be different. docs.google.com/document/d/1...

I challenged Gemini Antigravity to resurrect my programs, emails, and letters from almost 50 years ago. It re-implemented all but one of my old programs. Claude completed the set. docs.google.com/document/d/1...

I just read news.stanford.edu/stories/2026... from Stanford.

With Claude I created a large app (7500 lines of code) for exploring symbiosis in evolution. Sadly, I got negative, uninteresting, results. docs.google.com/document/d/1...

With the help of Codex I created a simulation of the 1918 Influenza Pandemic along with learning materials and an ask-AI feature. In 2012 creating the simulation took me almost half a year, this new model took me half a day. docs.google.com/document/d/1...

I made several enhancements to the local model chatbot app I created with Gemini last month. But Gemini let me down when I tried to add more models and features. I succeeded after switching to Claude. docs.google.com/document/d/1...

I continued my discussion with Claude after reading Corporations Constituting Intelligence. Besides getting a better understanding of AI governance issues, I was surprised to see so many introspective statements from Claude. docs.google.com/document/d/1...

I just read Public Constitutional AI, a paper by a Yale Law professor. I had lots of ideas and questions and then proceeded to have a very good, deep, conversation with Claude about how to govern AI. docs.google.com/document/d/1...

After reading “A Lo-Fi Rebellion Against AI” in the New Yorker I wondered how hard it would be to prompt ChatGPT and Gemini to generate images that don’t have an AI look. Chatbots thought many of the images might be mistakenly thought to be by people. docs.google.com/document/d/1...

I played with a new tool from Anthropic that let me (dimly) see the inner thoughts of an LLM as I asked it to write poetry, create an app, and answer questions about my book. docs.google.com/document/d/1...

I scanned a paper I wrote over 40 years ago and asked Codex to make an HTML version of the paper where the reader can run and modify all the examples in the paper. I’m happy with the result. docs.google.com/document/d/1...

I returned to the app I co-created with ChatGPT a few weeks ago where the Logo turtles draw on 3D shapes. I added cubes and several new features. docs.google.com/document/d/1...

With Gemini’s help I built an app that periodically creates illustrations in real time based upon what is being said. youtu.be/96jN2OCOfLs?...

Claude helped me build an animated message to aliens that begins with turtle graphics and ends with science and mathematics. Inspired by the 1960 book on the Lincos message. docs.google.com/document/d/1...

I asked ChatGPT whether a hack that enables programs to use the OpenAI API without a costly API key could be done in web pages as well. API usage would then count against the model usage quota of the ChatGPT subscriber. It is possible but very clumsy and hacky. docs.google.com/document/d/1...

After reading "What Will It Take to Get AI Out of Schools?" in the New Yorker I wondered how Seymour Papert would react. So I asked 3 chatbots to imagine response essays by Seymour. I was impressed. docs.google.com/document/d/1...

I returned to experimenting with the evolution of Logo turtle programs. This time I co-created with Claude a huge app (7000 lines) that automates experimentation and analysis. docs.google.com/document/d/1...

Today I created an app with Gemini and ChatGPT where a child can talk to unicorns, hear them “sing”, and give them presents. docs.google.com/document/d/1...

All the major chatbots now can include interactive visualizations inside their chats. I decided to challenge them with the complex process of how ribosomes create proteins. docs.google.com/document/d/1...

Logo turtles can escape flatland and leave trails on spheres, cylinders, cones, and donuts. With ChatGPT I built an app that runs Logo programs on 3D shapes. docs.google.com/document/d/1...

Last week, I created an app with an agent that modifies Logo turtle programs and another agent who evaluates the generated images and suggests improvements. I returned by working with Claude to improve the app and run over 75 experiments. docs.google.com/document/d/1...

I was able to design and implement a chatbot with support for vibe coding by asking chatbots to make the app run small free large language models. It runs fine without network access. docs.google.com/document/d/1...

After seeing a 1913 crossword puzzle in the Guardian today I asked ChatGPT to make an app from the photo. Its response was a working app. docs.google.com/document/d/1...

I spoke to ChatGPT on my phone to turn it into a spirit level for hanging pictures. Permissions to use its motion sensors complicated what should have been a simple task. docs.google.com/document/d/1...

With Gemini’s help I created an app with an agent that modifies Logo turtle programs and another agent who evaluates the generated images and creates suggested improvements based on the “taste” input. I ran experiments where the agents evolved the turtle programs . docs.google.com/document/d/1...

Yesterday in an online course based upon my book, The Learner’s Apprentice, two school students presented impressive apps they co-created with AI. One was a 3D space game made by a 9 year old and the other an “everything” app for fellow students. docs.google.com/document/d/1...

I guided Claude and Gemini in the creation of Seymour Papert bots. Some of the Papert quotes they made up, so I managed to get Codex to carefully go through 22 papers to select good thought-provoking quotes to start the conversations. docs.google.com/document/d/1...

I built playful mockups of a version of the Mac Finder based upon animation principles as imagined in a 1988 paper. I asked three chatbots to create demos of this “Disney Finder”. docs.google.com/document/d/1...

Today I easily built an extension with Claude that alters the fonts on web pages to match the age of the words. docs.google.com/document/d/1...

I created a Lemmings-like game where players control creatures by entering rules in English. Claude helped me brainstorm the design and implemented the code. GPT Codex tested the game and worked with Claude to improve it. docs.google.com/document/d/1...

I created over a dozen artifacts based on my book, The Learner’s Apprentice, using Google’s NotebookLM. I created these in nine different formats (videos, quizzes, podcasts, …). NotebookLM is impressive. docs.google.com/document/d/1...

A 5-year old’s adventures as a unicorn who meets fairies, a mermaid, and an initially very friendly shark by using an app I co-created with Gemini. docs.google.com/document/d/1...

I found that GPT-5.3-Codex can design, implement, test, and debug a multimedia game for a 5-year old. My job was mostly giving it permissions. docs.google.com/document/d/1...

I recently asked different chatbots to create novel web games. Here I asked them to create games that are both original and FUN. 10 games of varying quality resulted. docs.google.com/document/d/1...

After discovering a lack of novelty in yesterday’s LLM designed games I chatted with Claude about this. We came up with a 5-step prompt that worked well. docs.google.com/document/d/1...