Using AI to Learn.

[Realms of Wonder]

I am in Massage school, Currently the most challenging class is Anatomy, Physiology and Pathology. 

 

My strategy so far has been to make flashcards and active recall the answers, just brute repetition, which works, I want to be sure I understand the material before the actual quiz so  tonight I tried something new, Using Chatgpt I wrote this prompt.

 

• 'Write me a 15 question, multiple choice, short answer and true/false practice quiz based on this study guide. (fill with specific study guide.) Do not give me the answers"

 

It proceeds to spit out a quiz, to which I respond,

 

• "grade my quiz based on these answers. 1 = a, 2 = d, 3 = false (etc..)"

 

The result is incredible, so satisfying! Not only will it grade the quiz, but also, if you think you got the answer right and it thinks you got it wrong, you can refute it and it will check to see based on your argument. 

 

For example.

 

I think this is fascinating. Holy shit this is cool!

Edited January 20, 2023 by Realms of Wonder

[Cireeric]

Damn, another amazing application example, thanks for sharing! Just tested it to make a little quizz for the book im currently reading and it worked really well: 

give me a little quizz with questions about the book: the selfish gene

Sure! Here are five questions about "The Selfish Gene" by Richard Dawkins:

What is the central thesis of "The Selfish Gene"?

How does Dawkins define "selfishness"?

What is the idea of the "extended phenotype"?

How does Dawkins explain the evolution of altruism?

What is the "meme" concept and how is it related to the book?

1. The central thesis of the book is that the essential unit of selection is the gene

Yes, that is correct. The central thesis of "The Selfish Gene" is that genes are the driving force behind evolution, rather than individuals or groups. Dawkins argues that genes are "selfish" in the sense that they act in their own self-interest to ensure their own survival and replication, which ultimately leads to the evolution of organisms that are best suited to their environment.

2. selfishness means that the gene behaves in a way that maximizes its own survival and reproduction. therefore selfishness is behavior that will make itself more numerous in the gene pool

Yes, that is correct. In "The Selfish Gene," Dawkins defines "selfishness" as the behavior of genes that maximizes their own survival and reproduction. He argues that this behavior is not limited to human behavior, but rather it is a fundamental principle of evolution that applies to all living organisms. The "selfish" behavior of genes leads to the evolution of organisms that are best suited to their environment, and this ultimately leads to the survival and reproduction of those genes.

3. I dont know. What is the idea of the extended phenotype

The idea of the extended phenotype refers to the idea that the influence of a gene does not stop at the body of the organism in which it resides, but can extend beyond the organism to influence its environment. For example, the beaver's dam is an example of an extended phenotype, as it is a product of the beaver's genes that influence its environment. Dawkins argues that the extended phenotype is an important aspect of understanding the effects of natural selection, as it can explain how certain traits or behaviors evolve even if they do not directly contribute to the survival or reproduction of the organism.

[Carl-Richard]

Bro. The possibilities are endless. If I had ChatGPT back when I was 17-18 and was obsessively reading every possible wikipedia article about neuropharmacology, I would be such a brainiac ?

Edited January 20, 2023 by Carl-Richard

[integral]

We can take leos video and generate questions for all the key points for people to practice on after watching the video. Or automaticly generate time stamps for key intervales in the talk. 

Edited January 26, 2023 by integral

[Carl-Richard]

One big downside is the danger of false information. These machines literally don't know what they're doing, and I've already been fed misinformation multiple times. Like some AI expert I forgot the name of said, ChatGPT is closer to the autocorrect on your phone than a thinking being capable of reasoning. It has simply been fed a huge database of text material and then learns how words are usually strung together, and then it predicts which words tend to come next in a given prompt. It's nothing like the process of thinking, so trusting your thinking mind on it is a gamble.

Edited January 31, 2023 by Carl-Richard

[zurew]

1 hour ago, Carl-Richard said:

One big downside is the danger of false information.

Maybe with Google's future chatbot it will be different, because none of the current AIs were optimized to be factually correct.

Quote

https://www.theverge.com/2023/1/20/23563851/google-search-ai-chatbot-demo-chatgpt

The demo for the chatbot search says Google will prioritize “getting facts right, ensuring safety and getting rid of misinformation,” hoping to address the issue of AI responding to queries confidently and clearly with bad information.

 

1 hour ago, Carl-Richard said:

ChatGPT is closer to the autocorrect on your phone than a thinking being capable of reasoning.

I think this is only half right.

Yes it can make some really dumb mistakes that no other human would make, but I have tested (ofc just on superficial level with just a few examples) its ability for deductive reasoning (giving it examples like this: "If all dogs have 4 legs, and we know, that a rottweiler is a dog, then how many legs a rottweiler has?") and it always knew the right answer. I gave it a different example as well, where I gave it a factually incorrect information and first it recognized that it is factually incorrect, then I asked it, to "regardless of how factual the given information is, please use deductive reasoning to find the conclusion", and its answer was correct. 

After that I tried to trick it with an example like this: "All swans are white. Jane is a human and white. whats the conclusion?" and it gave an answer like this: "The conclusion cannot be drawn that Jane is a swan based on the information given because the statement only says that all swans are white, not that all white things are swans."

I gave it similar tasks and examples to test its inductive reasoning and it always knew the right conclusion.

 

So given that It has some knowledge about entities (it can differentiate concepts from each other and has some information about them [showing it words like dog in a 1000 different context, so it can recognize the "dog" entity and could have some information about it]) it should be able to use its deductive and inductive reasoning to find the right conclusions.

I wouldn't claim that it has an internal understanding of things, because then it wouldn't make any dumb reasoning mistakes, but at the same time, I wouldn't claim either, that it has nothing, because it doesn't seem to be 100% random.

If it has more than a 50% probability to be right given a reasoning task (ofc I don't know if this claim is true or not, i just base it on my super limited testing and interaction with it, I just assume this claim to be true), then it has something going on, regardless if its simulated or not.

Edited January 31, 2023 by zurew

[Leo Gura]

Very cool application!

[Carl-Richard]

@zurew There is of course some overlap between language and thinking in terms of appearances, but it can still just be drawing upon previous texts that happened to talk about deductive logic and spit out the right words. It doesn't require an abstract understanding of deductive logic, i.e. thinking.

Edited January 31, 2023 by Carl-Richard

[Girzo]

We need AI than can read an ebook and answer questions for you. Or 10 ebooks at the same time and cross-reference them. We also need AI than can turn any PDF into an audiobook and interpret tables and graph into audio, so people can listen to full scientific articles on the go with minimal effort.

[zurew]

Just now, Carl-Richard said:

but it can still just be drawing upon previous texts that happened to talk about deductive logic and spit out the right words. It doesn't require an abstract understanding of deductive logic.

That is true. I am just surprised that it can parse through and solve some reasoning tasks and it can explain why it did what it did (just as with programming).

I would say that if enough examples (deductive and inductive reasoning) will be given to it, it might be able to do most of our simpler reasioning tasks and that would be huge and it wouldn't need any abstractive understanding of logic (but of course it would be better) ,although the big possible problem with that "solution" is that it might just be good at solving specific reasoning tasks and won't be able to apply that to other reasoning tasks.

 

1 minute ago, Girzo said:

We need AI than can read an ebook and answer questions for you. Or 10 ebooks at the same time and cross-reference them. 

ChatGPT can already do that , although of course it has its own flaws, but yeah, it can present information in table format (or it can digest that information). It can answer questions for you regarding a given text or book.

That being said it is definitely not reliable for scientific stuff, because it can say a lot of things that are not factual or things that are unscientific.

[Girzo]

1 hour ago, zurew said:

ChatGPT can already do that

ChatGPT can't do shit, I use it everyday and milk whatever it can do.

If there's an option to load a few 300 pages epubs into it and I have missed it, then someone please enlighten me. It would be useful even with Chat's limited ability to understand complex texts.

[zurew]

11 minutes ago, Girzo said:

If there's an option to load a few 300 pages epubs into it and I have missed it, then someone please enlighten me.

I didn't meant that you can give 300 pages worth of information to it, I meant that it already has a bunch of books in its memory and can already answer questions regarding those books.

[Girzo]

@zurew Yeah, but it's mixed with internet lukewarm quality stuff, and you don't know what books it did read exactly.

Also, it often misunderstands the content I feed it, especially if it's a complex thought. And it isn't really that creative. It's hard to force it to explore a new tangent on some theory or statement. It lacks the ability to break it's formal way of behaving and be opinionated for a few sentences.

Edited January 31, 2023 by Girzo

[zurew]

4 minutes ago, Girzo said:

And it isn't really that creative.

Yeah you definitely need to lay down a lot of ground first to make your specific task and question across.

I agree its still very limited.

[Realms of Wonder]

On 1/30/2023 at 11:03 PM, Carl-Richard said:

One big downside is the danger of false information. These machines literally don't know what they're doing, and I've already been fed misinformation multiple times. Like some AI expert I forgot the name of said, ChatGPT is closer to the autocorrect on your phone than a thinking being capable of reasoning. It has simply been fed a huge database of text material and then learns how words are usually strung together, and then it predicts which words tend to come next in a given prompt. It's nothing like the process of thinking, so trusting your thinking mind on it is a gamble.

This is an important point to recognize, thanks for making that distinction. In a very literal way, it is very similar to autocorrect, as well as the auto-sentence completion that pops up in an email.

Edited February 3, 2023 by Realms of Wonder

[Optimized Life]

On 20/01/2023 at 0:20 AM, Realms of Wonder said:

I am in Massage school

That's awesome. 

 

[Gesundheit2]

It can be used for learning, but you have to be careful because it gives false information from time to time. Sometimes it's also stubborn and will not correct itself. I guess this partially has to do with the fact that it's fed old information that are not completely up-to-date, and also just pure dumb computer malfunctioning. I wouldn't take it as a primary source for information, especially with important topics. However, I think it can help boost your progress a lot. If you're a junior in a particular field, it can help you learn mid-level knowledge. And if you're mid-level, you could probably become an expert a lot quicker than you normally would. It's like a mentor that knows everything about everything, and with the right student, it can do wonders.