Compliment, Not Replace

Hey human,

Happy October half-term for those that celebrate it. For those that are still slogging, there’s not long to go now. Here’s your breakdown as to what’s new with AI and education.

📚 AI+education news

• AI Tutors in School > According to TES, the DfE are actively looking at how schools can use AI tutors (I previously reviewed one). It now looks that there will be explicit mention of how schools can be encouraged to look at using AI tutors. If it helps schools, I have previously recommended that a good Intelligent Tutoring System should consider these seven concepts:

  1. Expertly crafted curriculum

  2. Scaffolded learning

  3. Active recall and spaced repetition

  4. Interleaving for conceptual understanding

  5. Encouraging problem generation

  6. Error correction and reflection

  7. Cognitive load management

  The government are keen to reflect in their message that this is designed to complement the work teachers do, not to start replacing them.

• AI Coaches in School > Schools Week reports that the National Institute of Teaching is actively looking at AI-supported coaching. Lessons would be recorded to a tool that then converts it to a transcript. According to the report, these will be analysed by experts who will help train a specific AI model. Again, this is not to replace the human-to-human coaching that would take place, but to rather complement it.

• LaTeX in NotebookLM > An update that is sure to get Craig Barton excited. Now that it supports LaTeX, this means that the accuracy of the mathematical equations it can render, will be much more accurate.

🌍 Wider AI updates

• Identify Yourself > A new law in California now mandates that a Chatbot must identify itself as AI and not human.

• AI Presenter > Channel 4 recently ran a programme called Will AI Take My Job. Spoiler alert: The presenter herself was a creation of Generative AI, both her image and her voice. It was pretty convincing if I do say so myself.

🎯Prompt

• Unlocking Diverse Answers > Sometimes asking for diverse answers from an LLM can give you some interesting avenues to explore that you may not have thought of before. A new paper examined how to avoid mode collapse (where the LLM, likely through human training, prefers a typical response rather than an atypical one) by including the following phrase at the start of the prompt, ‘Generate X responses and their corresponding probabilities‘. X is the number of responses you want. Give it a go next time you want multiple responses to a query.

Till next week.

Mr A 🦾