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.
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.
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.
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.
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.
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.
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?
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 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.
The practical guide to using AI to do stuff
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.
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.
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.
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.
Thanks for being a teacher to whom has never had the privilege to sit in your classroom.
H
Have you tried you.com. It combines AI tools and social networks above with search. Just wondering what you think about this approach.
A weeks work here in following these leads up. And it would be a week very well spent!
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.
Great article, what does the 'AI' entitled diagram at the end depict(To pixelated to make out any detail when enlarged)?
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
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.
Using it to draft non-sensitive emails and legal documents has been a massive time saver for me
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?
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.
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.