What have I learned from Barbara Oakley’s Oxford lecture

Yesterday morning I was on the fence whether to go to Oxford to attend Barbara’s lecture. Both the train fare and time commitment are not negligible.

So help me decide whether it was worth it.

First of all, some quick stuff that you might want to check out right away.

Good books to read about GenAI


  • The worlds I see
  • the alignment problem


  • Co-intelligence (Ethan Mollick)
  • Teaching with AI, a practical guide to a new era of human learning (Jose Antonio Bowen)

Recommended Coursera courses

Generative AI for Everyone

Navigating Generative AI: A CEO Playbook

Generative AI Essentials: Overview and Impact

Introduction to Artificial Intelligence (AI)

Recommended places to find AI tools

FutureTools Collects & Organizes All The Best AI Tools So YOU Too Can Become Superhuman

Discover the best AI tools with Supertools.

Tools mentioned by Barbara

Scite for AI powered research

Semantic Scholar also for AI powered research

What I have learned

Lesson 1 Try different models

I have always found ChatGPT to be the best for me and dabbled Claude, Gemini, and llama 3. I did notice the difference in performance, however I still believed that ChatGPT is a good enough to serve as my go-to model and didn’t really try to use other models in my daily life. For me, the UI and functionalities (custom GPT, custom instruction) are just better.

However, Barbara made me realize that I might be wrong. I might be missing out a lot of interesting things by sticking to only one model and it is against the general wisdom that the more models, the better the overall result, though in this case, information overload can be an issue. I highly suspect that one is enough for most use cases but anyone should adjust the mentality that any single model is enough for other models.

Lesson 2, teach with metaphors

Metaphor is the best way to teach and learn. Research has revealed that the top teachers and lectures are successful because they leverage the power of metaphors. It is the case because metaphors can ease learning and good hooks can spark curiosity, which activates learner’s brain.

How to find the best metaphors? Use GenAI. GenAI is great at coming up with good and creative metaphors, such as this one:

This is actually very similar to what I had learned in a Coursera course that I took a while ago. The instructor, Jules White, showed how he used ChatGPT to come up with ketogenic and Kazakstani recipes, a rare combination. ChatGPT did a good job at it. With ChatGPT, creativity can be augmented as it is really easy for ChatGPT to connect the dots (due to its architecture I think) of various things and create something. We can put two different things, far away from each other in terms of semantic space, and produce brand-new ideas. And that is why LLMs are so good at producing metaphors.

In addition, Barbara also introduced the concept of working memory. Human brain can hold at the same time a maximum of 4 items (2 to 8 depending on the research). LLMs, however, can hold millions of items to create connections and new ideas.

Lesson 3, future generations should learn GenAI, no matter what

Barbara share that she told her grandchildren to broaden their interest, explore broadly, but make sure to learn GenAI.

Lesson 4, Barbara’s personal story

It is never too late to learn something new. Barbara was not good at math but she got a degree in engineering later in her life and “changed her brain”.

I have always wanted to learn a third language, programming. I should go for it.


Do you think it was worth it?

I think it was totally worth it. I will continue to learn GenAI, read the books and take the Coursera courses. It is great. I am so happy that I live in this age.

Next time when I cannot decide whether to go to an event, the default should be:

Just do it.

If you also want to listen to the lecture, the recording is here:


Leave a comment

Leave a Reply