26 Comments

Ethan - I appreciate many of your comments on AI. With respect to coding, I think that there are number of issues that are being ignored at the moment that make it less useful and more dangerous than you think. I don't know how much coding experience you have and therefore what assumptions I can make in this conversation.

There are a couple of things about writing software that are often misunderstood.

1. We spend most our time writing code. We don't, we spend much more time reasoning about the problem trying to understand it. In addition code is written once and read many times. So the cost of writing code is an order of magnitude smaller than the cost of deciding what code to write and rereading the same code later.

2. Since code must be correct (it solves the problem), secure, readable and efficient - our AI coding tool has just changed the cognitive burden. Instead of spending time writing code, the reciepent of this code needs to spend time reasoning "Is this code correct? Secure? ...?". I find it harder to reason about code someone else wrote than code I write myself.

I think there will yet be value in AI coding tools, it will be more of the nature of:

- Critique this piece of code for security, readability, ....

- I have a series of automated tests (proof of correctness) - write code that solves the problem required by the tests. _That still misses the other issues if security, readability, etc._

So AI useful for coding, just not yet and definitely not in the way you claim.

Expand full comment

RE: Coming Up With Ideas

Give Seenapse a try (https://seenapse.it). Seenapse is a unique combination of the ingenuity of human lateral thinking and the speed and automation of artificial intelligence.

Expand full comment

Amazingly helpful Ethan. How much did you use ChatGPT in writing it? That aside, I really like the vibe of embracing AI, rather than running from it. This has the feel of the heady days back in the 1990s when people were waking up to the web. But I think you rightly have included some warnings, e.g. hallucinating AI. Keep writing, I'm really enjoying it.

Expand full comment

Extremely helpful and I hope over time you can update this and share the updates as you and your students learn more. There are a lot of rapid changes happening here, and I will be sharing your insights with some friends as well as using it myself.

One point I include in discussing this area is that one should also never trust one internet site to have the "truth" on any subject, but rather learning is a continual process of checking various sources and asking good questions. This applies to new AI tools and everything else I do with the internet.

Expand full comment

Thanks for being a teacher to whom has never had the privilege to sit in your classroom.

Expand full comment

H

Expand full comment

Have you tried you.com. It combines AI tools and social networks above with search. Just wondering what you think about this approach.

Expand full comment

A weeks work here in following these leads up. And it would be a week very well spent!

Expand full comment

In the creative process I like to use the AI to iterate my own ideas / concepts. By inputing my own art or text, I can output multiple variations, exploring the possibility space I wouldn't be able to do on my own. It works a bit like a (very non-judgemental) buddy for bouncing ideas around. Quite incredible.

Expand full comment

Great article, what does the 'AI' entitled diagram at the end depict(To pixelated to make out any detail when enlarged)?

Expand full comment

How about the research process? Productivity and depth of insight will increase exponentially, especially for non-experts and young learners: https://medium.com/@stanfordgseit/a-new-class-of-ai-tools-part-2-ai-boosted-research-1fc3a107e70b

Expand full comment

Really appreciating these articles. Love this idea of fully embracing these technologies and encouraging their use. Students and others are smart folks. Trying to ban them has the potential to make issues worse.

Off to try the AI image generator for a logo. Love that idea. Great stuff.

Expand full comment

Using it to draft non-sensitive emails and legal documents has been a massive time saver for me

Expand full comment

Really helpful and interesting, thanks! My sense is that going forward if we educators want students to write and we want to be able to grade them or help them learn to write better, we can assume they'll use AI and encourage them to do so, and then go from there. Banning or curtailing its use seems counterproductive. Not only is that going to be impossible to police but what is the point of seeing what they can do entirely on their own vs what they can do when building on/working with AI?

Expand full comment

Your post was very informative (as always).

I would like to add the following.

When we speak, we use several different languages at the same time.

For example, when I am talking to a Bengali

in Bengali, I will often use a mixture of Bengali

and English. When I talk to my wife, we often

use many Spanish words or phrases that are

critical to the sentence structure. A monoligual

person would not get it. We used to use that

in front of our grandchildren when we did not

want them to *really* understand what we

were saying. Kids are always listening even

though you think they are engaged with some

other activity.

That led me to the following question:

If I asked GPT a question with a mixture,

what would it do?

Here is my question: Pythagoras Theorem

আমাকে বুঝিয়ে দেবে? (Can you explain it to me?)

GPT: Pythagoras Theorem states that in a right-angled triangle, the square of the length of the hypotenuse (the side opposite the right angle) is equal to the sum of the squares of the lengths of the other two sides. It can be written as c^2 = a^2 + b^2, where c is the length of the hypotenuse and a and b are the lengths of the other two sides. This theorem is named after the Greek mathematician Pythagoras, who is credited with its discovery.

This tells me one thing: When they used

the training data, they did not put different

languages in different silos.

That implies, the correlation matrices in the

training sets did not force off diagonal elements

across different environments to be zeros.

It would be tempting to do so with sparse

matrices. It would cut down the training time.

But it would also mean that we would get a

silo effect.

Executive summary: Scarily impressive.

Expand full comment

This is excellent, many thanks! Just what I wanted to know, all in one place. The all in one place part is quite helpful. I've been reading about these tools, but didn't have it together enough to organize what I'd learned. Thanks much for the spoon feeding.

Expand full comment