Joining a growing number of states, California’s Department of Education released “Guidance for the Safe and Effective Use of Artificial Intelligence in California Public Schools” recently. Positioned as advisory suggestions rather than required rules, the document stands out for its focus on the two lenses of Learning with AI and Learning about AI, as well as its attention to equity and access. The Golden State’s AI guidance is also unique in framing the issue of urgency as it relates to future skills that students will need to be successful.

Data from EdWeek’s Research Center, shows that 61% of teachers now report using AI tools in some part of their jobs, compared to 34% when asked in 2023. A big reason for the shift is an increase in funding for, and availability of, professional development for educators. The same data showed that in 2025, 50% of teachers received at least one PD session about AI. Another explanation is the nearly ubiquitous embedding of AI capabilities into the types of tools that teachers were already using, such as Quizlet, Canva, and MS Office.

Continuing their rivalry with back-to-back new product announcements, both OpenAI and Anthropic released new tools that put autonomous agents in the hands of everyday users. Subscribers to the premium plans for ChatGPT and Claude can now give extremely complex tasks to their AI and the models will engage multiple agents to breakdown the objective, complete separate components and put together the final deliverable, even making decisions along the way. In a viral post, AI startup CEO Matt Shumer has tried to help casual LLM users appreciate how big this change is.

Most teachers that I work with at this stage have heard about Google’s NotebookLM tool and it’s ability to create podcast-like Audio Overviews with real-sounding synthetic voices discussing resources. But there is so much about this tool that many educators have not yet explored.

NotebookLM is a type of RAG artificial intelligence tool, which stands for Retrieval Augmented Generation. This means that instead of just mixing it’s training data with your prompt when you give it instructions, it uses specific resources that you choose and includes those in the context window whenever it is given a prompt. You have much more control over the source materials and therefore much less hallucination takes place.

But beyond this capability, as Vera Cubero has reported, there is another powerful function: personalities. You can already give your LLM of choice directions (called “custom instructions”) about how you want it to respond to you. Many frontier chatbots also have a memory function that learns from your feedback to improve the quality of the output that it provides. However, those tools only allow one “personality” at a time. No matter where you use ChatGPT (on the web, in the app, etc) it will always have same personality on your account.

NotebookLM, however, allows you to customize the way it responds in each of the “notebooks” you create. This unlocks a new way to customize your chatbot. For example, I have a “just the facts” straightforward personality on my Notebook that acts as an F.A.Q. about my course. But in the Notebook that I use to collaborate as I build slide decks for presentations, I ask NotebookLM to ask for clarification and to lean toward constructive criticism.

How do I get started?

Vera calls these “personas” and provides some examples and detailed instructions in her DeepDive NotebookLM guide. While NotebookLM’s Terms of Service disallow its use by those under 18, it can be a real boon for teachers. One of my favorite (and most used) Notebooks inspired by Vera’s work uses these instructions to create it perfect persona:

Try these out in NotebookLM and see how the structure responses and concrete source materials elevate your output and get you closer to your ideal product more quickly.

Next week, I’ll be sharing the Learning Forge philosophy at the Science for All Summit and next month at the North Carolina Technology in Education Society Conference. If you’ll be attending either of these, I’d love to meet and chat about how you’re using AI in your classroom. You can use this link to check out my slides after February 19.

This spring will also bring more of Solution Tree’s “AI for Educators” workshops all over the country. I’ll be kicking it off in Stockton, California but other dates will be listed soon. Check out ST’s website to sign up or to schedule an event near you.

In March, as I celebrate my 50th birthday, I’ll be sharing some of the feedback from my presentations. Until then, keep looking for ways to be curious and try AI tools in new ways.

Paul (and the Codium Educational Consulting team)

P.S.

Don’t forget to email me with examples of how you’ve used the tools and strategies that I’ve shared.