AI Enhanced Education

[IFR]

I'm going to declare here that I'm impressed with the state of AI tools, and in fact now find them invaluable and use them on a nearly daily basis.

I particularly like GPT 4 with its plugins. There are fantastic plugins available, such as one which allows GPT to make queries to Wolfram Alpha, and one which allows it to read pdf documents.

I generally like to play outside of my own field in my readings, and some of the material is very challenging without any aid. GPT has fundamentally changed that.

It is obviously an imperfect system. AI LLM have a widely known problem with hallucinations, and so it's very important to verify information and evaluate everything critically. But used properly, I'm finding this to be an unbelievably powerful tool, and I'm very excited to see the future improvements, and the effect it may have on education in general.

What does everyone here think?

Edited June 21, 2023 by IFR

[Starkess]

I think it's dangerous to rely on something that will blithely and convincingly lie to provide an education on any topic.

[IFR]

49 minutes ago, Starkess said:

I think it's dangerous to rely on something that will blithely and convincingly lie to provide an education on any topic.

That is a fair take. I can only speak anecdotally, and as someone who is in a pretty advantageous position to be surrounded by experts in various fields, but so far I've encountered very few situations where I've been misled. I make an effort to corroborate information, and part of the reason I'm so impressed is that this tool has been surprisingly reliable.

As a counterpoint, I consider people who parse text on their own without an AI tool. I think it's likely that their own initial understanding, without an AI aid, is probably going to be far more misdirected.

That is to say, I'm not suggesting that AI is anything other than an imperfect tool. But I think pursuing knowledge without its benefit is even more imperfect.

[maarsen]

People have been parsing text since text was invented. AIs don't understand the written word so what is the point? Read the article in the latest Scientific American for a description of the problem. 

[IFR]

1 hour ago, maarsen said:

People have been parsing text since text was invented. AIs don't understand the written word so what is the point? Read the article in the latest Scientific American for a description of the problem. 

I looked at the article you refer to. It simply went into the fact that much of the processing of a given input into these LLM architectures lacks transparency and can be a black box. It then describes the approach of transformer networks, that use a network of probabilistic weighting factors to eventually churn out an output.

And my question to you is, so what? Why does it matter whether the tool understands what it's doing?

Let me illustrate by example. Say I'm trying to understand the mechanics of a Geiger-Muller tube. As a neophyte, I don't know what I'm doing right? I pick up a textbook, say I'm lucky and get Radiation Detection and Measurement by Glenn Knoll, which happens to be a good one.

I, the neophyte, see this passage:

Quote

In a typical Townsend avalanche created by a single original electron, many excited gas molecules are formed by electron collisions in addition to the secondary ions. Within usually no more than a few nanoseconds, these excited molecules return to their ground state through the emission of photons whose wavelength may be in the visible or ultraviolet region. These photons are the key element in the propagation of the chain reaction that makes up the Geiger discharge. An energetic optical photon emitted in the filling of an inner electron shell vacancy may be reabsorbed elsewhere in the gas by photoelectric absorption involving a less tightly bound electron, creating a new free electron. Alternatively, the photon may reach the cathode wall where it could release a free electron upon absorption. In either case, the newly created free electron will migrate toward the anode and trigger another avalanche (see Fig. 7.1).

Now I actually understand this quite well. But if I was neophyte then a lot of this may be confusing.

Let's see how GPT breaks this down:

Quote

 

This passage is describing the process of a Townsend avalanche and its role in a Geiger discharge, which is a type of electrical discharge used in a Geiger-Müller counter, a device for detecting ionizing radiation.

Here's a step-by-step breakdown of the passage:

1. **Townsend Avalanche**: This is a process that occurs in a gas subjected to a high electric field. It starts with a single electron that, due to the electric field, gains enough energy to ionize a gas molecule when it collides with it. This ionization event creates additional free electrons.

2. **Excited Molecules and Photon Emission**: The ionization process leaves gas molecules in an excited state. Within a few nanoseconds, these excited molecules return to their ground (normal) state. The energy difference is released in the form of photons, particles of light. These photons can have wavelengths in the visible or ultraviolet region of the light spectrum.

3. **Photon Role in Geiger Discharge**: These emitted photons play a crucial role in the propagation of the Geiger discharge, which is the chain reaction of ionization events that make up the functioning of a Geiger-Müller counter.

4. **Photon Absorption and Electron Release**: The photons emitted by the excited molecules can be reabsorbed in two ways. They can be absorbed by other gas molecules, causing photoelectric absorption. This process involves a less tightly bound electron absorbing the photon's energy and becoming a free electron. Alternatively, the photon can reach the cathode (negative electrode) and release a free electron upon absorption.

5. **New Avalanche**: In either case, the newly created free electron will move towards the anode (positive electrode) due to the electric field. This electron can gain enough energy to cause another ionization event, triggering another Townsend avalanche. This chain reaction is what allows the Geiger-Müller counter to detect ionizing radiation.

The "Fig. 7.1" mentioned in the passage likely refers to a diagram illustrating this process.

 

So far, so good. But perhaps I still don't understand parts of this. I could spend a bunch of time looking through wikipedia, where the articles are primarily written for the highly technical audience in mind, or I could google through countless intro science pages - which comes with its own landmines of misinformation. Or I could continue to prompt GPT.

Here we go. What is ionization?

Quote

 

Ionization is the process by which an atom or a molecule acquires a negative or positive charge by gaining or losing electrons, often in conjunction with other chemical changes. This results in the formation of ions.

There are several ways ionization can occur:

1. **Collision with a charged particle**: This is the most common form of ionization and is used in many types of particle detectors. An atom can be ionized by a collision with a high-energy charged particle, such as an electron or a proton. The charged particle can transfer some of its energy to one of the atom's electrons, giving it enough energy to escape from the atom and creating an ion.

2. **Absorption of electromagnetic radiation**: If an atom absorbs a photon (a particle of light) with enough energy, one of its electrons can gain enough energy to escape from the atom, resulting in ionization.

3. **Chemical reactions**: Some chemical reactions can result in ionization. For example, when an acid dissolves in water, it ionizes to produce hydrogen ions.

The resulting ions can have positive or negative charges. A positive ion, or cation, is produced when an atom loses one or more electrons. A negative ion, or anion, is produced when an atom gains one or more electrons.

 

How about why an excited electron emits a photon?

Quote

 

When an electron in an atom absorbs energy, it can move from a lower energy level to a higher energy level, a state we refer to as being "excited." However, this excited state is not stable, and the electron tends to return to its original lower energy level, or "ground state."

The energy difference between the excited state and the ground state is released when the electron transitions back down. This energy is emitted in the form of a photon, which is a particle of light. The energy of the photon corresponds to the energy difference between the two states. 

This process is known as "spontaneous emission." The emitted photon's energy determines its frequency or color. For example, if the energy difference is large, the emitted photon might be in the ultraviolet range. If the energy difference is smaller, the photon might be in the visible or infrared range.

This phenomenon is the basis for many forms of light emission, including the glow of neon lights and the colors seen in a fireworks display. Each element has a unique set of energy levels, and thus each can emit light at only specific frequencies, giving each element its unique color signature.

 

In this process of prompting GPT, there's a probability of occasionally being exposed to misinformation. But first of all, this is a very finely simplified evaluation of dense text. And secondly, reading is no longer a passive experience, but an active experience.

So I find it hard to be persuaded that this utterly useless because of the problem of hallucinations. When people learn new information, invariably there will be some degree of misconceptualization in the process. Chances are, if you tried to read that text cold, you would on many levels miss, or misunderstand, what was being said. Students do this frequently with this text.

Can AI be improved? Absolutely. It's still in its infancy, and there's a long ways to go. But even in this form, it seems very useful to me.

[lacuna]

So to benefit from AI, I have to get the AI to explain [thing], then I have to read up on [thing] via other sources so that I can verify that the AI isn't feeding me bullshit? Why go through AI then?

[IFR]

53 minutes ago, lacuna said:

So to benefit from AI, I have to get the AI to explain [thing], then I have to read up on [thing] via other sources so that I can verify that the AI isn't feeding me bullshit? Why go through AI then?

I think this is a good question. My perspective is this: First, it makes reading an active process. You are always in conversation, instead of just absorbing static information.

Secondly, I find that a dynamic process of learning is helpful, personally. Maybe the initial explanation doesn't really make sense. You can search the internet for different kinds of explanations on the same topic, and perhaps this will help. But the power of AI tools is continuous interactive feedback. I can immediately tell the AI that their explanation doesn't make sense to me, and that it should simplify further. I can ask them to provide different examples. I can put their explanation in my own words and see if I'm missing anything. The point is not to memorize information, but to build an in-depth, generalized model in your head that makes sense to you. Which takes effort, of course. And you can't expect to have experts at hand to spend all day explaining the nuance of everything to this point. But after you have a sufficiently solid model in your head, then you can test it for how well it holds up.

So there may be some percent of inaccuracies in the sequence of responses the AI provides. But, I found personally, that in an effort to build an overall model, the inconsistencies become apparent where they exist, and that is when you want to examine more critically where you've gone wrong and why. And this - at least in my experience - is pretty easy to do, because you now have a pretty good sense of the big picture. And so it actually is a  quite efficient avenue of learning. And, of course, if still doesn't negate access to other educational tools (it encourages their use, in fact).

And this is the state of AI now. It is probable in the future the problem with hallucinations will be minimized, likely to a high degree. In which case this process of dynamic learning will become even more efficient.

[Proudfeet]

Think generative AI is overhyped. It looks most useful simplifying menial processes, but for other uses, without the ability to actually verify the authenticity, its just a poor search engine which only gives one result.

As I said in the other thread on AI generated books, its a flying car. Sounds great, but is actually hard to implement and even harder to create. Just another waste of time and money after cryptocurrency and metaverse.

[IFR]

3 hours ago, Proudfeet said:

Think generative AI is overhyped. It looks most useful simplifying menial processes, but for other uses, without the ability to actually verify the authenticity, its just a poor search engine which only gives one result.

As I said in the other thread on AI generated books, its a flying car. Sounds great, but is actually hard to implement and even harder to create. Just another waste of time and money after cryptocurrency and metaverse.

Possibly. I'm glad I can make use of it, but you may be correct that it isn't sufficiently advanced for a broader audience. Time will tell.

Anyway, even though others do not share my enthusiasm, I do appreciate the opinions in this thread. 

Edited June 22, 2023 by IFR

[Theda Baratheon]

I haven’t really been using any AI tools other than a little dabbling with starryai to make some AI art for myself. 
 

What exactly would be the point of using an AI tool to write a job application? Wouldn’t it first require a lot of time and effort to input all of your history anyway. So would you even be saving any time? Surely companies will start building their own counter AI tools to detect this sort of thing? 

[dog-days]

Detecting AI: Think it could be quite difficult to do given that AI doesn't copy text. It builds content based on likelihood, so follows patterns. Which sounds to me a lot like how humans write. 

I've got my job history stored on my CV and on LinkedIn.  If I was less lazy I'd also have recorded details about what those jobs in involved. 

Given that material, then with some careful formulation of the task, I guess you could get a decent supporting statement out of ChatGPT. 

Edited June 27, 2023 by dog-days

[Theda Baratheon]

9 minutes ago, dog-days said:

Detecting AI: Think it could be quite difficult to do given that AI doesn't copy text. It builds content based on likelihood, so follows patterns. Which sounds to me a lot like how humans write. 

I've got my job history stored on my CV and on LinkedIn.  If I was less lazy I'd also have recorded details about what those jobs in involved. 

Given that material, then with some careful formulation of the task, I guess you could get a decent supporting statement out of ChatGPT. 

I think using it to write a cover letter or something like that could make sense. A but of work in the short term for less in the long term. I am not really the most efficient person I must admit lol. 

[Buckwheat]

On 6/22/2023 at 12:42 PM, IFR said:

You can search the internet for different kinds of explanations on the same topic, and perhaps this will help. But the power of AI tools is continuous interactive feedback. I can immediately tell the AI that their explanation doesn't make sense to me, and that it should simplify further. I can ask them to provide different examples. I can put their explanation in my own words and see if I'm missing anything. The point is not to memorize information, but to build an in-depth, generalized model in your head that makes sense to you. Which takes effort, of course. And you can't expect to have experts at hand to spend all day explaining the nuance of everything to this point. But after you have a sufficiently solid model in your head, then you can test it for how well it holds up.

A teacher can do all of that. As well as ask you questions to make sure you understood the material and can explain it in your own words, use humour or anecdotes to make the material easier to learn and understand, as well as encourage you to work together with other students of the same topic ...

(Need to defend the role of teachers here.  )

Edited June 28, 2023 by Buckwheat

[IFR]

5 hours ago, Buckwheat said:

A teacher can do all of that. As well as ask you questions to make sure you understood the material and can explain it in your own words, use humour or anecdotes to make the material easier to learn and understand, as well as encourage you to work together with other students of the same topic ...

(Need to defend the role of teachers here.  )

 It's true, but in my experience a teacher's time is not inexhaustible, nor is their patience. And sometimes their explanations are not put in terms that make things clear (not through lack of effort, perhaps, it's just that certain approaches may "click" with the student, and the teacher does not find that right special approach).

There's a radiopharmaceutical  production program that uses a linear accelerator to produce a radionuclide that can be used for cancer targeted therapy and diagnostics. It's still in it's clinical stage. I'm acquainted with many of the people involved with this program, which happens to be pretty niche at the moment, especially with this radionuclide.

I was given a binder of the research papers used to parameterize the production methodology. Some of this was outside my area of knowledge.

I slowly went through the research papers and queried GPT 4 in depth on parts that were ambiguous to me. It was a very exciting and educational process of learning in real time. An actual expert by my side who could explain every passage that I read would have been much better, but no one has the time for that.

I would have had a decent understanding of everything without AI, but I think AI not only made the reading much more efficient and enjoyable, but also furthered my depth of understanding of the project. I spent about a day on this undertaking.

After developing a broad but in depth understanding of the project, I then consulted with the people on the project. I knew exactly what questions to ask, and when they clarified details, it really made sense. This made my role in all of this significantly easier.

[kayrich]

On 6/21/2023 at 8:28 PM, IFR said:

I'm going to declare here that I'm impressed with the state of AI tools, and in fact now find them invaluable and use them on a nearly daily basis.

I particularly like GPT 4 with its plugins. There are fantastic plugins available, such as one which allows GPT to make queries to Wolfram Alpha, and one which allows it to read pdf documents.

I generally like to play outside of my own field in my readings, and some of the material is very challenging without any aid. GPT has fundamentally changed that.

It is obviously an imperfect system. AI LLM have a widely known problem with hallucinations, and so it's very important to verify information and evaluate everything critically. But used properly, I'm finding this to be an unbelievably powerful tool, and I'm very excited to see the future improvements, and the effect it may have on education in general.

What does everyone here think?

I think AI is for sure will change education soon, but I'm not sure whether those changes will be positive or negative. Or maybe, either. I've tried using ChatGPT for different reasons and can say that it works, but it's still not a perfect tool. But it's still in development, and who knows what will be possible to do with it in the future.
But I know that now it's very popular among students, but I think it's visible when a student uses it, for example, for writing. The professor will see the difference because they know the level of the student, and if the task is too good, it sure means that some help was used. Sometimes I also use help when it comes to writing, but I use https://edubirdie.com/capstone-project-writing-service, and I can provide all the requirements I have, explain some details, so the finished paper will look perfect. And I've never faced any problems besides such writing, which I use as an example. So that when I practice my skills, I can compare them all, make some conclusions, and work on my mistakes.

Edited October 29, 2023 by kayrich

[mcbigski]

Neal Stephenson covered this in his third book or so diamond age.