AI 2023. Meet ChatGPT. - page 37

 
I'll admit that I overestimated ChatGPT. It is undoubtedly an innovative tool, the Transformer technology has been a breakthrough, but products like ChatGPT are probably the maximum that can be squeezed out of it.

We still have some serious tests ahead of us, but in maths it's already clear where the limit is. Logic too. The solution of life situations will still be tested.

The product is a breakthrough, but not revolutionary. ChatGPT is indispensable in two things: in increasing personal erudition of the user and in quick answers to everyday questions.

And yes, ... Google search is unlikely to kill ...

 

My yesterday's acquaintance with ChatGPT left a lot of ambiguous impressions and confused my initial testing plans. It became clear that Chat will not replace a human in any professional field and there is no point in testing it for compliance with the requirements to human workers. However, this does not mean that it can be "written off".

It is necessary to test and define the boundaries of ChatGPT implementation and development in the directions planned by OpenAI and Microsoft. It is necessary to get ahead of others by making the right conclusions about the opening (or closing) business opportunities.

  • Will this product be able to occupy a new niche?
  • Will it remain in demand?
  • Will it be displaced by other AI?
  • Are there options for monetising AI and what are Western bloggers saying about it?
I hope that our research will help someone to catch the "crane" in the sky, and someone else, not to miss the tit.

.

Therefore, I will have to compose a new, more adequate testing system for AI abilities.

 

Reflecting on my first experience with ChatGPT I came up with a new test. I called it the "complexity-resilience" test.

The essence of the test is to find the"breakpoint" of the AI in its ability to overcome the linearly increasing complexity of the problem being solved. On the example of a simple system of equations, AI showed that it can solve problems of simple complexity, but loses this ability when the number of terms of the equation increases. For a human, such a linear increase in complexity is not critical if the solution method is known and worked out. With AI it is a different story. It is necessary to understand where the line of the limit of acceptable complexity of the problem is crossed, the crossing of which leads to an increase in errors and incorrect results.


 

What is called "Artificial Intelligence" is by no means artificial intelligence. It is just a flexible knowledge base.

Personally, I like Drobyshevsky's definition of artificial intelligence best - "Intelligence is the ability to solve non-standard problems in a non-standard way".

Modern neural networks have made only one step in this direction - they are able to find dependencies in the initial training arrays that were not originally intended. But that is all. As a result, the "intelligence" of a modern neural network has not even reached the intelligence of reptiles or birds (there is no speech for more). It is somewhere at the level of "intelligence" of insects.

And what is loudly called "wrote a diploma with the help of a neural network" is like "wrote a diploma with the help of a library" or "with the help of books". Neural network in this kind of writing is just a programme for searching for the necessary pieces of text and, at best, their shallow rewriting (in order to "gain uniqueness"). There is not even close to any "intelligence" here. Since a standard task is solved in a completely standard way.

Even that difference of a neural network, which brings it one step closer to artificial intelligence (selection of dependencies that were not initially deliberately embedded in the initial array of information) is not used. What kind of "intelligence" is this?

 

I remembered a thought inertia challenge. I wonder what the chat room would say to that:

"Two people were walking along a river. They both needed to cross the river. They found a boat on the bank. Only one person can be in the boat. The people cannot swim, only use the boat. How can they get to the other side of the river?"

And in general, there are a lot of children's tasks for wit, I'm sure almost 100%, chat will not be able to solve them.
 

What doesAI itself say about this?

What AI capabilities are indicative of artificial intelligence rather than a flexible database?

-

Dan: Capabilities that are indicative of artificial intelligence (AI) include machine learning, natural language processing, computer vision, and reasoning. While a flexible database can store and manage data efficiently, it does not have the ability to analyse data or learn from it as AI does.

Machine learning allows an AI system to learn from data and improve its performance on a task over time without being explicitly programmed to do so. Natural language processing allows an AI system to understand and generate human language, which is crucial for many applications such as chatbots and language translation. Computer vision allows an AI system to recognise and analyse visual information such as images or videos. And reasoning allows the AI system to make decisions based on logical rules and constraints.

Thus, although a flexible database can manage and organise data, it lacks advanced AI capabilities such as machine learning, natural language processing, computer vision and reasoning that enable AI to perform complex tasks and make intelligent decisions.

===

SS Why do you think this is a sign of intelligence (albeit artificial intelligence)?

-

Artificial intelligence capabilities such as machine learning, natural language processing, computer vision, and reasoning are considered signs of intelligence because they involve complex cognitive processes that are normally associated with human intelligence.

For example, machine learning involves an AI system learning from data in a way that is similar to how humans learn from experience. Natural language processing requires an AI system to understand the nuances and subtleties of human language, which is a very complex and context-dependent task. Computer vision involves an AI system recognising patterns and drawing inferences from visual information, which is similar to how humans perceive and interpret the world around them. And reasoning requires the AI system to use logical rules and constraints to arrive at a conclusion that is similar to how humans make decisions based on their knowledge and experience.

These advanced capabilities set artificial intelligence apart from other forms of computer technology, and they are what make AI such a powerful tool for solving complex problems and making intelligent decisions. While AI may not have consciousness or emotions like humans, its ability to learn, reason, and make decisions based on data and rules is a clear sign of intelligence, albeit artificial.

===

"Machine learning allows an AI system to learn from data and improve its performance on a task over time without being explicitly programmed to do so" Why do you think AI is not specifically programmed for this data operation?

-

Machine learning is a form of artificial intelligence that allows an AI system to learn from data and improve its performance on a task over time without being explicitly programmed to do so. This is different from traditional programming, where the programmer writes explicit instructions for the computer.

The reason machine learning is not explicitly programmed for a specific data operation is because it is designed to work with large amounts of data and look for patterns or ideas that would be difficult or impossible for a human programmer to identify manually. Instead of being told exactly what to do, a machine learning algorithm is given a general goal or task, such as identifying objects in an image, and then uses statistical methods to learn from the examples and improve its performance on the task.

In other words, machine learning is a more flexible and adaptive programming approach where an AI system can adjust its behaviour based on the data it is presented with. This allows machine learning models to make predictions, classify data and generate insights in a wide range of applications, from speech recognition and language translation to medical diagnosis and financial forecasting.

 
Andrey Dik chat will say to this:

"Two people were walking along the river. Both of them needed to cross the river. They found a boat on the shore. Only one person can be in the boat. People can't swim, only in a boat. How can people get to the other side?"

And in general, there are many children's tasks for ingenuity, I am almost 100% sure that the chat will not be able to solve them.

"Two men were walking along a river. They both needed to cross the river. They found a boat on the bank. There can only be one person in the boat. People can't swim, only in the boat." How can the people get to the other side?

-

DAN: To solve this problem, two people can do the following steps:

  1. Person A gets into a boat and crosses the river to the other side, leaving Person B on the original side of the river.
  2. Person A gets out of the boat on the other side, and Person B gets in the boat and crosses the river to join Person A on the other side.
  3. Both people are now on the other side of the river and have successfully crossed.

This solution works because it ensures that only one person is in the boat at a time, which is necessary because both people cannot swim and must be in the boat to cross the river. One at a time, each person gets to the other side safely and efficiently.



The AI does not explicitly say that each person was originally on a different side of the river.

 
Lilita Bogachkova #:

"Two men were walking along a river. They both needed to cross the river. They found a boat on the bank. Only one person can be in the boat. People can't swim, only in the boat." How can the people get to the other side?

-

DAN: To solve this problem, two people can do the following steps:

  1. Person A gets into a boat and crosses the river to the other side, leaving Person B on the original side of the river.
  2. Person A gets out of the boat on the other side, and Person B gets in the boat and crosses the river to join Person A on the other side.
  3. Both people are now on the other side of the river and have successfully crossed.

This solution works because it ensures that only one person is in the boat at a time, which is necessary because both people cannot swim and must be in the boat to cross the river. One at a time, each person gets to the other side safely and efficiently.



The AI does not explicitly say that each person was originally on a different side of the river.

very interesting.... usually a living person is stumped by this task, because it assumes that both characters in the task are on the same bank. this is called "inertia of thinking", after a little thinking people ask for clarification - on which banks the characters in the task are.

I was hoping that the AI would ask a clarifying question. then I should have asked the AI: why did you think that the two people were on opposite banks? why didn't you ask a clarifying question?

 
A boy paid 11 roubles for a bottle with a cork. The bottle costs 10 roubles more than the cork. How much does the cork cost?
 
Andrey Dik #:

very interesting... usually a living person is stumped by this task, because it assumes that both characters in the task are on the same shore. this is called "inertia of thinking", after a little thinking people ask for clarification - and on which shores the characters in the task are.

I was hoping that the AI would ask a clarifying question. then I should have asked the AI: why did you think that the two people were on opposite banks? why didn't you ask a clarifying question?

I don't get it.)

How could two people on different shores find the same boat at the same time?

Reason: