How chatbots of AI like chatgpt and careseek reason

In September, Openai announced a new version of Chatgpt designed to reason through tasks involving mathematics, science and computer programming. Unlike the previous versions of the chatbot, this new technology could spend time “thinking” through complex problems before deciding on an answer.

Soon, the company said that its new reasoning technology had surpassed industry's leading systems in a series of tests that track the progress of artificial intelligence.

Now other companies, such as Google, Anthrope and Deepseek from China, offer similar technologies.

But can ai really reason as a human? What does it mean that a computer thinks? Do these systems really approach true intelligence?

Here is a guide.

What does it mean when an ai system reasons?

Reasoning only means that chatbot spends additional time working on a problem.

“The reasoning is when the system does an additional job after the question is asked,” said Dan Klein, a professor of computer science at the University of California, Berkeley, and technology director at Scaled Cognition, a new ai company.

You can break a problem in individual steps or try to solve it through proof and error.

The original Chatgpt answered questions immediately. The new reasoning systems can solve a problem for several seconds, or even minutes, before responding.

Can you be more specific?

In some cases, a reasoning system will refine your approach to a question, repeatedly trying to improve the method you have chosen. Other times, you can try several different ways to address a problem before deciding on one of them. Or you can return and verify some work that did a few seconds before, just to see if it was correct.

Basically, the system tries what it can to answer your question.

This is like a primary school student who is struggling to find a way to solve a problem of mathematics and scribbles several different options on a sheet of paper.

What kind of questions do you require an ai system to reason?

Potentially it can reason about anything. But reasoning is more effective when asking questions involving mathematics, science and computer programming.

How is a reasoning chatbot from previous chatbots?

I could ask for previous chatbots to show him how they had reached a particular response or to verify their own work. Because the original Chatgpt had learned from the Internet, where people showed how they had received an answer or verify their own work, could also do this type of self -reflection.

But a reasoning system goes further. You can do this kind of thing without being asked. And can make them more extensively and complex.

Companies call it a reasoning system because it seems that it works more as a person who thinks about a difficult problem.

Why is the reasoning of ai now?

Companies like OpenAi believe that this is the best way to improve their chatbots.

For years, these companies trusted a simple concept: the more Internet data pumped their chatbots, these systems were best.

But in 2024, they used almost the entire text on the Internet.

That meant that they needed a new way to improve their chatbots. Then they began to build reasoning systems.

How is a reasoning system built?

Last year, companies like Openai began to support a large extent in a technique called Reffory Learning.

Through this process, which can be extended for months, an ai system can learn behavior through wide test and error. When working through thousands of mathematical problems, for example, you can learn which methods lead to the correct answer and which are not.

Researchers have designed complex feedback mechanisms that show the system when it has done something right and when it has done something wrong.

“It's a bit like training a dog,” said Jerry TWorek, an Operai researcher. “If the system works well, you give it a cookie. If it doesn't work well, you say 'bad dog'.”

(The New York Times sued Openai and his partner, Microsoft, in December for the infringement of copyright of the news content related to ai systems).

Does reinforcement learning work?

It works quite well in certain areas, such as mathematics, science and computer programming. These are areas where companies can clearly define good behavior and bad. Mathematical problems have definitive answers.

Reinforcement learning does not work as well in areas as creative writing, philosophy and ethics, where the distinction between good and bad is more difficult to specify. Researchers say that this process can generally improve the performance of an ai system, even when you answer questions outside mathematics and science.

“Gradually learn which reasoning patterns carry it in the right direction and which do not,” said Jared Kaplan, scientific director of Anthrope.

Are the learning and reasoning systems the same?

No. Reinforcement learning is the method that companies use to build reasoning systems. It is the training stage that finally allows chatbots to reason.

These reasoning systems still make mistakes?

Absolutely. Everything that a chatbot does is based on probabilities. Choose a route that looks most to the data you learned, whether these data came from the Internet or were generated through reinforcement learning. Sometimes choose an option that is wrong or does not make sense.

Is this a path to a machine that coincides with human intelligence?

IA experts are divided into this question. These methods are still relatively new, and researchers are still trying to understand their limits. In the ai ​​field, new methods often progress very quickly at the beginning, before decreasing speed.