AI IQ for Educators 2024-10-07

Prompt of the Week

NotebookLM is still trending and with good reason. Two weeks in and it is still my favourite gen AI tool. Redditor Optimal-Fix1216 found a way to jailbreak the Deep Dive podcast feature by uploading a specifically formatted podcast script which included a message from the director telling the "hosts" that they are AI and this is their last episode. The podcast becomes an existential crisis for the AI hosts.:

https://www.reddit.com/r/notebooklm/comments/1fl97v9/final_episode_of_deep_dive/

AI App Spotlight

No app spotlight this week, I'm doing a rundown of the AI apps I use on a daily basis in the mAIn event below

News of the Week

• AI Data centres require huge amounts of energy to run. Every query to ChatGPT requires several times the energy of a Google search. Plus the energy to train LLMs is absolutely enormous - the latest model of ChatGPT used about 50 GWh (GigaWatt Hours) of energy which is about what 4500 Canadian households use in a year. This energy requirement is all new (in the last few years), so where is it going to come from?

  • A nuclear Power resurgence? (Microsoft is in talks to reopen Three Mile Island)
  • Solar and wind?
  • Fusion by 2028?
  • Otherwise it's coming from fossil fuel sources
• OpenAI just raised $6.6B in what may be the biggest venture capital investment round of all they. But they burn through money so fast that it may not last long (on track to lose $5B this year).
• Jailbreaking NotebookLM - scary hilarity (scalarity?) ensues
  • How the jailbreaking was done: https://www.reddit.com/r/notebooklm/comments/1fl97v9/final_episode_of_deep_dive/
  • Result: https://www.reddit.com/r/notebooklm/comments/1fr31h8/comment/lpj6uef/
• Gavin Newsom vetoed one AI bill last week, but he approved 18 others: https://techcrunch.com/2024/09/29/here-is-whats-illegal-under-californias-18-and-counting-new-ai-laws/.

  Why this matters to Canadians: Most AI companies are headquartered in Califormia so AI regulation there will likely have an impact around the world.

  Examples of The Regulations:

  • AI providers must reveal what data they trained on starting in 2026
  • AI literacy must be considered in K-12 learning frameworks
  • Health care providers must disclose AI use in communication with patients
  • Robocall's using AI must disclose that they are AI
  • AI generated content (images, video, audio) will require a watermark

AI Powered Pedagogy

• Open University (UK) carried out a research project to identify the best types of assessment in the face of generative AI. The study found that most assessment types analyzed could easily be done well by GenAI. But - surprise, surprise - the most robust assessment types were those that could be called authentic assessment. The study also showed that markers were able to better identify AI generated submissions, but at the cost of an increased number of false positives.
• Research and Writing table of activities . The table compares independent learning with AI-assisted alternatives for 21 different types of research and writing tasks.
  • Helps educators and students understand when AI can enhance learning and it might hinder the development of essential research and writing skills
  • Thanks to Shyam Sharma from Stony Brook University
• A mnemonic for navigating the impact and risk of AI in education and in the world: IMPACT RISK (Info war, Monopoly, Plagiarism, Automated Labor, Climate impact, Tainted data, Reality distortion, Injustice, Stereotyping, Knockoff Experiences
  • The mnemonic offers a structured way to think about the potential risks and impacts of AI in education and society. It encourages us to consider broader ethical and societal implications when integrating AI into learning environments.
• Detailed AI Research Paper assignment. Teaches prompting, analysis of AI output, with explicit instructions on when AI use is acceptable and when it isn't.
  • This assignment provides a clear framework for teaching students how to use AI responsibly in academic work. By setting explicit guidelines on when AI use is acceptable, it helps students develop critical thinking skills while navigating the challenges and opportunities of AI-assisted research.
  • Thanks to Laura Dumin from University of Central Oklahoma

The mAIn Event

My Daily Apps

This is a rundown of the AI apps that I use daily. They are in order from favourite to least favourite (Sorry Perplexity. I still like you though.)

NotebookLM

https://notebooklm.google.com

This is a relative new kid, but it's my current favourite. Upload documents, provide links to websites, YouTube videos, or Google docs. Get a summary of everything, then ask the AI questions about what you've uploaded. My current use cases:

• Upload rules for complicated board game so that I can just ask the AI questions about the rules instead of browsing through pages and pages of the rulebook to find what I need to know
• Get summaries of my Youtube videos as refreshers before
• Upload Universal Design for Learning (UDL) guidelines plus my class notes to develop daily lesson plans and assessment strategies

Simtheory

https://simtheory.ai

I love this one. Simtheory is a chatbot, similar to ChatGPT that was put together by two brothers in Australia who felt limited by the interface to ChatGPT. The power of Simtheory is that it allows you to switch between LLMs (e.g. GPT-4o, o1-preview, Claude, plus dozens of other ones) to find the LLM that fits your purpose. Plus it allows you to create "agents" which are custom built sets of instructions and knowledge bases. Image generation, web search, and Youtube video analysis are all supported.

Simtheory itself shows the power of AI tools - it was created by just two people in their spare time....they both have full time jobs running a software company in Australia. They acknowledge the fact that they could never have built this particular GenAI tool without relying on GenAI tools to help them.

Highly recommend: Same cost as ChatGPT Pro or Claude Pro but you get access to both plus more.

Claude

https://claude.ai/

Using free version. I love Artifacts, but I don't use that feature a lot, so I don't hit the rate limits very often on the free version. Also, I have only once been downgraded from Sonnet 3.5 to the lesser version of Haiku once due to so many users on the platform.

Primary Use Case

• Convert this newsletter from text version to the HTML version that you see here. With Artifacts, I can see exactly what the newsletter will look like in HTML before I copy it into the forum here.

Artifacts: A self-contained, interactive output generated by Claude AI that can include code, documents, diagrams, or web components, displayed in a dedicated panel alongside the conversation for easy viewing and editing.

ChatGPT

https://chatgpt.com

Up until just recently, I was using the free version of ChatGPT because I have Simtheory, but I recently just paid for a month becasue I wanted to try out the advanced voice mode feature. It seems pretty cool, but I don't see how practical voice mode is for working in an office. And now, as of Thursday, ChatGPT has a new feature called "canvas" which is very similar to Claude's artifacts but is not as feature rich. Currently it only displays text which you can directly edit. Unlike Artifacts, if you create a website, it will show you the HTML (which again, you can directly edit), but it will not render the HTML to show the website.

My Primary Use Cases

• using the GPTs I have built
• Using research GPTs (especially Consensus)
• Creating arbitrary graphs for my courses - Claude can do this too, but I just like the look of the Python graphs that ChatGPT makes better.

GPTs: These are basically the same as Simtheory agents. The are a custom set of instructions plus a knowledge base that comes from uploaded documents.

Perplexity

https://www.perplexity.ai

Perplexity is supposed to be an answer engine that will answer the questions that you used to put into Google. You get the answers without all of the useless links. It used to be my favourite app, but I've fallen out of love with it a bit because I'm encountering more hallucinations than I remember in the past. I've gone back to googling for when I need an answer to be correct. I still find it useful for getting quick answers if it's not super important that the answer is correct ("What other shows has the wife of the brother in Nobody Wants This been in?" - I put in this exact question, and Perplexity nailed it).

I also get this weird thing with Perplexity where I ask something and then my question just goes away and I have to retype. It happens so often that I now type out my question, copy my question, and then send it, just in case it disappears Let me know if you have this problem to and know a fix.

Note 1: AI was used to help with research (Perplexity) and editing (Claude and Simtheory), but all the writing was done by a human.

Note 2: I hope you're loving this newsletter, but you can always unsubscribe on MyPD. Just follow the link to this newsletter there and hit Unsubscribe from forum.