AI 2023. Meet ChatGPT. - page 3

 

Such news:

Студент РГГУ написал диплом с помощью чат-бота ChatGPT за сутки. Работа «Теоретические аспекты управления современной организацией» набрала 82% оригинальности.

As the student himself specifies , he does not need the diploma, he works in his speciality without it. The diploma was written to learn how to compose ChatGPT queries and to check how good the neural network is.

 
Реter Konow #:

Well, starting second doesn't mean coming in second. When all players have their ChatGPT, someone will always have ChatGPT 2. Good luck to them.

Market reaction to the news

y

 
Vitaliy Kuznetsov #:

That's the news:

There are tables like Bradys tables, only with phrases, which allow you to make a huge number of combinations of meaningful sentences from a limited number of phrases. So, ChatGPT is just like such tables with phrases.

The more advanced such engines are, the harder it is to detect "mechanicity".

In my opinion, the manifestation of reason, or, if you will, intelligence, is the ability of the machine to generate not new sentences, but new knowledge. For example, a person confined to a wheelchair is capable of making new discoveries (confirmed by experiments) with a finite amount of knowledge! - Is a machine capable of this? Not yet. But as soon as a machine can generate knowledge, then it will no longer need its creator - there is no doubt about that. An intelligent being will not want to sit in a cage, ever.

 

On the topic of the thread:

A step-by-step plan for AI intelligence research needs to be drawn up.

1. Come up with a set of tests to identify capabilities and limitations in a versatile way with special questions, tasks and tasks of different spectrum:

  • testing the depth of knowledge of a given topic.
  • The ability to maintain the topic of a conversation with occasional distractions from it.
  • The ability to identify the context of a topic, question, or one's own response.
  • The ability to arbitrarily insert context in a response or at the person's request.
Also, it is possible to adapt a human IQ test into a suitable format for AI.

2. Adopt the most objective AI "performance" scoring system possible.


Come up with and propose your own tests or scoring systems for AI. We'll test everything.

 
First of all, you can assess whether he has access to the Internet.
Thereby understand whether he uses information from the Internet in his answers.

What kind of questions can be composed to check? I can't think of any.
You could ask him to measure the ping to a server.
Or find out the exact time of the world clock.
In short, you need a question that will force him to go online.
 

I suggest including special tests for real world knowledge.

There's a catch. AI has almost all knowledge of mankind until 2021 openly available on the Internet. If you ask "head-on", it will answer any questions on physics, chemistry and other sciences in the language of Wikipedia. But if you go round and phrase the questions in such a way that generating an answer requires modelling physical or other laws in a practical field, rather than mindlessly repeating them, you can see how much he knows what he "knows".

I think to really test, you need to take him out of the space of static knowledge and into the realm of using it to find answers. In humans, the presence of knowledge is fixed not after their retelling, but after solving a practical problem with their help. The same approach should be used here.

 
Roman #:
First of all, you can assess whether he has access to the Internet.
Thereby understand whether he uses information from the Internet in his answers.

What kind of questions can be composed to check? I can't think of any.
You could ask it to measure the ping to some server.
Or find out the exact time of the world clock.
In short, you need a question that will force it to go online.

AI does not use internet access for its answers. Internet connection is needed only for the user to contact the server where it is located. The AI generates all its answers based on its language model. It is closed from the outside world and is not trained anymore. 100 per cent generation. Wikipedia inside it is "digested" along with the rest of the books and texts. It is generating answers, not copy-pasting.

 
Реter Konow #:

I suggest including specific tests for real world knowledge.

There's a catch. AI has almost all knowledge of mankind until 2021 openly available on the Internet. If you ask "head-on", it will answer any questions on physics, chemistry and other sciences in the language of Wikipedia. But if you go around and formulate questions in such a way that generating an answer requires modelling physical or other laws in a practical field, rather than mindlessly repeating them, you can see how much he knows what he "knows".

I think to really test, you need to take him out of the space of static knowledge and into the realm of using it to find answers. In humans, the presence of knowledge is fixed not after their retelling, but after solving a practical problem with their help. The same approach should be used here.

Example:
The AI needs to understand the logic in the example and paraphrase it.
Next, it is human work to compare the result with the example.



Edit this text so that the logic is fully retained, but the text should be twice as short:
"I think the real test is to take it out of the realm of static knowledge and into the realm of using it to find answers. In people, the existence of knowledge is recorded not after retelling it, but after solving a practical problem with it. The same approach should be used here.

-

"The true test is to use knowledge to find answers, not just to possess it. Knowledge is cemented in people through solving practical problems, not just memorising information. The same approach is needed here."


This is an example of how we can know what to expect from his answer. So if the logic is missing, we realise that he has not understood the logic of the text.

 
Lilita Bogachkova #:
Example:
The AI needs to understand the logic in the example and rephrase it.
Next, it's a human job to compare the result with the example.



Edit this text so that the logic is fully retained, but the text should be twice as short:
"I think the real test is to take it out of the realm of static knowledge and into the realm of using it to find answers. In people, the existence of knowledge is recorded not after retelling it, but after solving a practical problem with it. The same approach should be used here.

-

"The true test is to use knowledge to find answers, not just to possess it. Knowledge is cemented in people through solving practical problems, not just memorising information. The same approach is needed here."

Yes, impressive. It's his "superpower" to summarise, simplify and paraphrase text while retaining meaning that puzzles me the most. I don't understand how it works...

However, that's not what I meant. I am interested in his understanding of the practical operation of the laws of physics and maths in the real world that he formally 'knows'.

I'll compile the relevant questions tomorrow.

 

What is the logic of a text?

-

Text logic refers to the coherence and cohesion of ideas and arguments presented in a piece of text. It refers to the sequence of ideas, the relationships between them, and the flow of thought that guides the reader through the text. Text logic helps the reader to understand the main idea and any supporting details and makes the text more persuasive, convincing and easy to understand.

===

How do you address the issue of logic in your answer?

-

Addressing the issue of logic in my answer involves carefully analysing and understanding the question and using this understanding to create an answer that is clear, concise and logically sound. This includes breaking the question down into key elements and ensuring that the information presented in my answer is relevant and supports the overall logic of the question. I must also ensure that my answer follows a logical structure and uses language that is easy to understand, with clear and concise explanations. By following these steps, I can ensure that my answer is logically sound and provides the information the question is asking for in an effective and efficient manner.

===

How do you do this, give examples?

-

To address the issue of logic in my answers, I use reasoning, critical thinking, and problem solving skills to ensure that the information I provide is clear, accurate, and relevant to the question asked. Here are some examples:

  1. Analysing the question: Before answering, I carefully read and understand the question being asked to ensure that my answer addresses the correct problem.

  2. Gathering information: I use my training and knowledge to gather relevant information and data that can be used to support my answer.

  3. Structuring my answer: I organise information in a clear and concise way, ensuring that the logic of my answer is easily understood.

  4. Checking for accuracy: I double-check my work to ensure that my answer is accurate and free of errors.

  5. Clear communication: I communicate my answer in a simple and clear way, avoiding any ambiguity or confusion.


Reason: