How to look like AI like Chatgpt and Deepseek Mind

How to look like AI like Chatgpt and Deepseek Mind

by

In September, Openai unveiled a new version of Chatgpt designed to interpret it through tasks that include sports programming, science and computer. Unlike previous versions of Chatbot, this new technology can spend time “thinking” through complex problems before settling an answer.

Soon, the company said that the new thinking technology outperformed the leading industry systems in a series of tests that follow the progress of artificial intelligence.

Now other companies, such as Google, Anthropic and Deepseek in China, offer similar technologies.

But can Amnesty International actually cause human being? What does it mean to think of the computer? Are these systems really approaching real intelligence?

Here is evidence.

Thinking only means that chatbot spends some extra time working on a problem.

“Thinking is when the system does an additional work after asking the question,” said Dan Klein, Professor of Computer Science at the University of California, Berkeley, and the chief technology official at Scaled Cromed, AI, from Amnesty International.

It may divide a problem into individual steps or try to solve it through experience and error.

The original Chatgpt answered the questions immediately. New thinking systems can work through a problem for several seconds – or even minutes – before reply.

In some cases, the thinking system will improve its approaches to a question, and repeatedly trying to improve the way he chose. At other times, it may try several different ways to deal with a problem before settling on one of them. Or it may return and check some of the actions she did a few seconds before, to see if it is true.

Basically, the regime tries everything in his power to answer your question.

This is similar to the primary school student who is fighting to find a way to solve the problem of mathematics and preach many different options on a paper.

It can be a reason about anything. But thinking is more effective when they ask questions that include mathematics, science and computer programming.

You can ask Chatbots earlier to show you how they have reached a specific answer or to check their work. Since the original Chatgpt may be learned from the text on the Internet, where people have shown how they reached an answer or examined their work, this type of self -thinking could also be done.

But the thinking system goes further. These types of things can do without asking. You can do it in more comprehensive and complicated ways.

Companies call it the thinking system because it seems as if it works like a person who is thinking about a difficult problem.

Companies like Openai believe this is the best way to improve Chatbots.

For years, these companies relied on a simple concept: the more Internet data they have done in their Chatbots, the better the performance of these systems.

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

This means that they need a new way to improve Chatbots. So they started building thinking systems.

Last year, companies like Openai began to rely heavily on a technology called reinforcement learning.

Through this process – which can extend over months – the artificial intelligence system can learn behavior through a comprehensive experience and error. By working through thousands of mathematics problems, for example, he can learn methods that lead to the correct answer that do not do so.

The researchers designed complex reactions that show the system when he had done something right and when he made something wrong.

“It is somewhat similar to the training of a dog,” said Jerry Tourek, an Openai researcher. “If the system is good, then you give it a cookie. If it is not good, then you say,” a bad dog. “

(The New York Times filed a lawsuit against Openai and its partner, Microsoft, in December for violating copyright for news content related to artificial intelligence systems.)

It works well in certain areas, such as mathematics, science and computer programming. These are the areas that companies can clearly define good and bad behavior. Mathematics problems have final answers.

Reinforcement learning also does not work in areas such as creative writing, philosophy and ethics, as it is difficult to determine the distinction between good and bad. The researchers say this process can generally improve the performance of the artificial intelligence system, even when it answers questions outside mathematics and science.

“He gradually learns what patterns of thinking leads in the right direction and which of them,” said Jared Kaplan, chief science employee in Anthropor.

no. Reinforcement learning is the way companies use to build thinking systems. It is the training phase that ultimately allows the caratbots the mind.

definitely. Everything that Chatbot does depends on the possibilities. He chooses a path similar to the data you learned from it – whether this data came from the Internet or created through reinforcement learning. Sometimes he chooses a wrong or meaningless option.

Artificial intelligence experts are divided into this question. These methods are still relatively new, and researchers are still trying to understand their borders. In the artificial intelligence field, new ways are often advanced very quickly at first, before slowing down.

Leave a Comment