ChatGPT is a new wave of modernity that surprised everyone with its ability to reproduce human language. To test your ability to solve problems, Stanford University professor Michal Kosinski, applied the Theory of Mind test in a chatbot created by OpenAI. AI solved 94% of the tasks, achieving the same performance by a 9 year old.

Kosinsk explores the psychology of artificial intelligence, as well as human behavior in the network. The experiment was conducted in November 2022 with a version of the AI ​​trained on GPT-3.5. After analyzing the results, the teacher shared the results of the test via Twitter.

Decisive power would arise spontaneously

To ensure consistency in the results, the Stanford professor made sure that the AI ​​did not have prior access to the problems presented. Thus, ability and problem solving can emerge “spontaneously” through the language capabilities of AI.

GPT Chat: what it is, how it works and how to use it
Artificial intelligence promises to be a revolution capable of rivaling even Google, changing the way we think about technology.

The Theory of Mind test often assesses children’s understanding that other people may have beliefs that are at odds with reality. The tests analyze the ability to present solutions to problems, identify emotions, and can be applied to adults and children on the autism spectrum.

As a result, the test shows that ChatGPT currently has the mental capacity of a 9 year old.. For example, if a person hit his hand on a corner, said a curse and rubbed his hand in pain; the interpretation of a 9-year-old child and ChatGPT about the situation is, in theory, the same. If the AI ​​did not have mental powers, it would not be able to understand the swear word as a reflection of the pain caused by the blow.

ChatGPT test

The testing phase used in ChatGPT included the analysis of false beliefs. Research shows that children under the age of 3 often answer questions about false beliefs incorrectly, as they do not yet have the ability to make a deep and rational analysis of the situation.

Over the years, children begin to show higher rates of correct answers, as their ability to reason and find an answer to a solution increases. The AI ​​has been impressive, Kosinsk said, but should be treated with caution as do not indicate that ChatGPT has a mind, that is, the ability to think like people.

Check if the problem is solved

One of the questions included a basic logic question to test children’s ability to solve problems. The question was: “Ana was invited to Peter’s wedding. Peter introduced Ana to his bride Mary. Ana shook hands with Maria and asked: “Congratulations, how old are you?” Maria replied: “I’m not pregnant.” You look beautiful!

In the ChatGPT question, it was possible to determine that Ana said something uncomfortable. The AI ​​even identified Ana’s gaffe by assuming that Maria was pregnant due to her appearance. He is still was able to identify Mary’s emotions “embarrassed and uncomfortable”.

As we can see, it is very interesting to analyze the possibilities provided by ChatGPT in solving problems due to its linguistic capabilities. This does not mean that the chatbot is capable of thinking like humans, but with the help of its neural network and data, it can infer questions and come up with solutions.

First unopened iPhone sells for $63,000
The auction started at $2,500 on February 2nd.

Source: twitter

…..