"Did you get Deep Research for free as a demo?"

In partnership with

Welcome to AutomatED: the newsletter on how to teach better with tech.

In each edition, I share what I have learned — and am learning — about AI and tech in the university classroom. What works, what doesn't, and why.

Today, I answer 3 reader questions on topics big and small related to tech/AI and pedagogy.

📣 By the way, I just spun up a little website to explain the AI tools I’ve developed to help department course schedulers:

PS - There are only 2 more department slots available for this semester.

📢 Quick Hits:
AI News and Links

1. Google launched “Career Dreamer,” an AI-powered career exploration tool that helps people identify transferable skills and discover new career paths. It uses Gemini to help craft career identity statements, explore job possibilities, and draft application materials. This could be particularly valuable for career services offices and academic advisors working with students on their career pathways. (More info here.)

2. Google also unveiled an "AI co-scientist," a multi-agent system built on Gemini 2.0 that assists researchers in generating novel hypotheses and experimental designs. In real lab experiments, it successfully predicted new drug treatments for leukemia and proposed liver fibrosis targets that showed "significant anti-fibrotic activity." Perhaps most impressively, it independently rediscovered a novel bacterial gene transfer mechanism that had been recently found (but not yet published) by researchers at Imperial College London. Google is opening access to research organizations through a Trusted Tester Program, and you can sign up here.

3. In a piece parallel to mine from a few months ago on AI and prestige, Hollis Robbins argues that “it’s later than you think,” in the sense that universities must hurry to radically reimagine their value proposition as AGI emerges: "If universities cannot articulate in detail how their faculty exceeds AGI capabilities, what value are they offering to tuition-paying students?" She predicts universities will retain only faculty who either advance original research beyond AGI capabilities, teach sophisticated physical skills, or work with previously undiscovered materials, and she argues that the other factors that make higher ed valuable, like social skill development, aren’t sufficient to justify the existence of higher ed.

4. OpenAI released a major update to their Model Spec, the document defining how their AI models should behave.

❓ My Answers to 3 Reader Questions

The mailbag returns! Here are a few of the inquiries I’ve received in the last two weeks from subscribers and clients, as well as my answers.

Remember that you can share your questions and comments by responding to any of my emails — responses to any AutomatED email come straight to my inbox — or by submitting a comment on one of my polls.

✨1. “What do you think of using NotebookLM Plus as an AI tutor, rather than a custom GPT?”

The person asking this question noted that NotebookLM seems to offer better control than a custom GPT over the knowledge base that the LLM relies on in responding to user queries.

As I detailed last week, NotebookLM Plus's ability to share notebooks in "chat-only" mode while enforcing "Guide" style interactions is compelling for tutoring use cases. Students can engage with your curated course materials through AI-guided discussion, with the system staying reliably grounded in those sources.

And yes, in my testing, NotebookLM's current technology does a better job staying focused on its knowledge base compared to custom GPTs' deployment of retrieval-augmented generation (RAG); the latter requires significant prompt engineering and testing to get good performance, partly because it seems that OpenAI hasn’t improved custom GPTs in this dimension in quite some time.

However, I would recommend to not write off custom GPTs just yet, whether for tutoring or otherwise. Major updates are coming in the next few months from OpenAI that could change this calculus dramatically.

Let me explain, starting with a key piece of information recently shared by OpenAI.

2. "Did you get Deep Research for free as a demo?"

A reader asked this after my enthusiastic review of OpenAI's new research tool. She was (jokingly?) skeptical that anyone would be this positive about a $200/month service without some... incentives.

Fair enough! That $200 monthly charge is painfully real!

However, as I told her via email, despite writing about AI tools for years, I've never received a free account or demo from any AI company. (OpenAI, Google, if you're reading this... 👀)

But her skepticism highlights something important: I only praise tools I've thoroughly tested with my own money and time. I spend 4-10 hours a day working with AI tools, so it takes a lot to get me pumped. If I am enthusiastic, it’s because I really believe what I am saying.

Deep Research actually impresses me a lot, citation quirks and all. And, by summer, those citation issues I noted will likely be sorted out, with even more features added as OpenAI works to stay ahead of Google.

3. "What's the most common tool that your consultation clients ask for help with?"

I got this question in a recent 30-minute exploratory meeting with a curious subscriber.

My answer won’t come as a surprise.

As I develop more and more custom AI tools for clients (including my department course scheduler), I find that teaching professors benefit the most from tools that help them with their feedback processes.

There are two reasons for this.

1. By and large, professors have always disliked (hated?) grading and feedback tasks, unless they are working one-on-one with a student in real-time. These tasks are time-consuming and draining because they are cognitively demanding and it often feels like you’re sending your feedback out into the abyss.

(Side note: this is one reason why I love doing and advocate for oral exams with live feedback, when feasible.)

2. LLMs are hard to efficiently and ethically use out-of-the-box for grading and feedback.

The solutions I build vary by context, but they often involve creating workflows that let professors focus on what they can uniquely provide — expertise-dependent evaluation —while AI handles everything else about generating their feedback. This avoids data privacy issues as well.

I have one client for whom I’m developing a series of tools to help with two different assignment types. For one assignment type, the professor used to spend 30 minutes per student submission, including time inputting everything into their LMS. They now spend 10 minutes per submission. With 40 students per semester and 5 such assignments, that’s 4,000 minutes saved per semester. I charged them $1250 for this tool, which amounts to ~$18.75 per hour saved. Yet, since they plan to use it for 3+ semesters with the same teaching load, that rate goes to $6.25 per hour at most. As they said to me, their time is worth a lot more than $6.25 per hour (indeed, it’s worth a lot more than $18.75!).

This brings up a longer discussion, but a crucial fact about this client is that they are paying for my services out of pocket. This isn’t uncommon either — most faculty don’t have funds from their universities for stuff like this. However, it’s my view that their institutions should be developing scalable solutions like this for them, not expecting them to either spend the time or the money themselves.

📝 Save the Date: ($0) Webinar This Friday

Want to learn more about on using AI to generate course content?

If so, just answer “Yes” in the below poll and you are thereby registered for my February Zoom webinar.

It will be this Friday, February 28th from noon to 1:30pm Eastern.

All registrants will receive the recording afterwards, even if they don’t attend live.

The rest of my webinars this semester will have an entrance fee, so this is your chance to attend without spending a dime.

If you select “Yes”, that’s enough to register. You’ll get a follow-up email this afternoon with more information, including the Zoom link.