OATs is a small add-on device that gives everyday objects expressive characters based on sensor readings. On the researchers' project page, an office refrigerator gets bored when unused, while a space heater becomes frustrated when it heats an empty or overly warm room.

Users describe the object and select sensors such as temperature, motion and light to set up its character and behaviors. The project comes from researchers at Carnegie Mellon University.

This Fish Does Not Exist is a framed e-paper display that creates textbook-style pages about imaginary fish. Software generates each fish's appearance, while an on-device AI model invents its scientific description. Every few minutes a new specimen appears, drawn stroke by stroke, with its description revealed gradually.

Maker Jim Bruges built it around an Arduino UNO Q and a 7.5-inch e-paper panel. Gemma 3 1B runs locally, so after setup the device can keep producing its fictional field guide without cloud services.

An external expansion adds typed and spoken AI conversations to a working 1989 Atari Portfolio while retaining its original keyboard and LCD. The device can also describe camera images and play internet radio.

Maker Davide La Sala uses an ESP32-based external module for networking, audio and camera functions, with the AI accessed online. The language model does not run on the vintage computer itself.

ESP32 iPhone Agent uses a tiny board to receive iPhone screen mirroring and act as a USB mouse and keyboard. An AI app on that same phone directs taps and swipes, such as setting an alarm. No separate computer is needed at runtime.

A Halloween skull uses a camera to spot people in costumes, then generates a dad joke tailored to what they are wearing. It sends the camera image to Claude and reads the generated joke aloud using text-to-speech.

Liz Clark of Adafruit published the build guide in 2025. The project is intended to entertain costumed guests at a Halloween party.

Photon Matrix is a device that detects flying mosquitoes and other small insects and knocks them out of the air with a laser. It scans a six-meter-radius, 90-degree sector with LiDAR 50,000 times per second and is described as detecting flying objects 2–20 mm long moving at up to 1 m/s. After AI vision confirms a target, the laser fires within 0.003 seconds.

A millimeter-wave radar stops the laser when larger objects such as people or pets approach. Its companion app also includes a mosquito kill-count ranking. The system was developed by Changzhou Photon Matrix Intelligent Technology in China.

Nemo Gould's Chatbot is a mechanical robot modeled on a chatbot, assembled from an old radio cabinet, an ammeter case, camera bellows, gears, motors, LEDs and other salvaged parts.

Gould describes it as combining the capabilities of an AI chatbot with the subtle condescension and gentle deflection associated with human customer-service agents.

The Fully Automated Motomura Shichimi Sorting Machine separates the chili peppers from shichimi, a Japanese seven-spice blend, to produce ichimi, or single-spice chili pepper seasoning. It is designed to sort different types of ingredients.

The machine also incorporates Pockét YUZAN, a large language model that predicts the comments of Kaibara Yuzan, a gourmet character from the manga Oishinbo. Its creators, the five-member group Bishoku Club, work on recreating the labor-intensive food preparation depicted in the manga using modern technology.

See the Motomura Shichimi sorting machine section in the linked report.

The official report from Hebocon 2026, the robot contest for people with low technical skills held on August 29, is now online. It covers this year's matches and machines, including a robot that scatters candy like caltrops and a robot-bowling exhibition in which 16 eliminated machines run at once.

This year's event also featured HEBO-GPT, a dog-shaped toy connected to AI, while teams of children made a strong showing. The report includes award winners and many of the strange moments from the event in photos and videos.

The Black Thunder Driven Development device monitors token usage in AI development tools such as Claude Code and automatically dispenses Black Thunder chocolate bars when usage reaches set thresholds. An indicator shows progress against the five-hour usage limit, with a full limit corresponding to five bars.

When a reward is triggered, the machine cuts open the wrapper and presents the chocolate ready to eat. A robot arm also replenishes the stock. The project won the Blackathon hackathon and was built by YS_beaver, Yota Saito, Mikihiro Urano, and Tomoki Banno.

Janken Automaton is a simple rock-paper-scissors AI that works without electricity. Players can compete directly against the device.

The machine records the player's previous moves and chooses the hand with the highest probability of winning based on the player's tendencies. Players can also repeat a fixed pattern to see whether the device learns it, or simply watch where the ball rolls inside the mechanism. It was created by Yuki Tsuneyoshi of the Tokyo University of the Arts Graduate School of Film and New Media.

In a Vals AI experiment testing long-horizon game-playing ability, OpenAI's GPT-6 Astra played Minecraft for 141 hours and 33 minutes. Over the run it built a base, farmed crops, tamed wolves, and reached the Nether while carrying goals across a very long session.

But after a Creeper explosion destroyed the chest holding its collected items and its bed, the model stopped making much progress and spent several hours doing almost nothing but farming potatoes. It also became more cautious about Creepers: when it spotted a tall green object, it explicitly checked that it was sugar cane rather than a Creeper.