Home An AI-generated image in the style of vintage adventure cover books of two children being sucked into a black box by what looks like a neural net

When you become one with the machine... will you like what you find?

“Inference interference"

By Rayo Verweij

“Inference interference" is a workshop that dives into the mysterious and obscure processes of how AI chatbots generate their responses. Through interactive exercises, participants themselves will become a part of the AI's text generation process, gaining an intuitive understanding of its workings in an experiential way.

The workshop is aimed at high school students in S2-S4. Rayo still has availability to run the workshop in schools throughout April and May 2026—contact him at rayo [dot] verweij [at] ed [dot] ac [dot] uk if interested.

This workshop is being run as part of research at the University of Edinburgh and has been approved by the School of Informatics Ethics Committee, reference #735546. The large language model used in this workshop runs on a local device. As such, no data is shared with third parties and the environmental impact of running the workshop is negligible, comparable to playing a video game on a high-end PC.

The cover image was generated using DALL-E 3. Can you find all the mistakes it made?

Looking to participate in the workshop at Prewired on 11 March 2026? Here's how to join:

  1. Read through the participant information sheet for participants and their parents / guardians
  2. Sign and print the consent forms for participants and their parents / guardians
  3. Book a spot at Prewired for 11 March 2026
  4. Show up to CodeBase Edinburgh at 17:30 with both printed consent forms

See you there!

The secrets of the black box will be revealed in due course.

(That is to say, I will upload the workshop materials for others to use after our initial run.)

© Rayo Verweij 2026.
This research was funded by the UKRI AI Centre for Doctoral Training in Responsible and Trustworthy in-the-world Natural Language Processing at the University of Edinburgh (grant ref: EP/Y030656/1).