The Chatbot Chatgpt was launched in November 2022, and its evolution has been enormous in its few more than two years of life. San Altman has told some secrets of his birth. Among them, that the first version was very bad chatting, and many Openai employees thought they would make a fool of them.
The OpenAi CEO has granted an interview to the media Stratecherywhere he has talked about The Origins of the Company, and Chatgpt. He has also recognized a thing he was worse than Grok. «But we have improved a lot in that,» he acknowledges.
According to Altman, Chatgpt’s first version was bad by maintaining a conversationbut they didn’t want to wait for GPT-4 because «that would basically overcome Turing, and people were not prepared for it.»
This was the origin of chatgpt
The Openai co -founder tells in the interview that the company was born through many meals in restaurants with the founding members, among which was also Elon Musk.
At first his idea was Create a research laboratoryeverything was so new that they had no goal, or a business program. A group developed an AI to play video games, and another to manage a business. There was no coordination, or goal.
When they developed a large language model (LLM), such as GPT-1 and 2, they began to realize what they could get with artificial intelligence.
With GPT-3 they already had an API, and the developers began to create applications that used AI. There was a tool called The Playground where they could try things in their air. And discovered that Many people tried to talk to AIalthough it was not designed for it.
«We had been thinking about a chatbot for a long time. Even before the API, we had talked about creating a chatbot, but we did not believe that the model was good enough,» says Altman.
It was when they were developing GPT-3.5 and GPT-4, when they realized that they could design a chatbot capable of maintaining a conversation as a human. And then they faced a great dilemma.
There was a debate in Openai: Some wanted to launch the chatbot with GPT-3, and others launch it as well as GPT-4which was the true revolution of the LLM: «We had all got used to GPT-4, version 3.5 was no longer so good, and there was much fear that if we launched something with 3.5, nobody was going to care and people were not going to be impressed.»
The OpenAi CEO made the decision to launch two products separately. ChatGPT in November 2022, and GPT-4 at the beginning of 2023.
This saved their lives, because Chatgpt was a revolution, and if they had launched it with GPT-4, which is a much more demanding language model at AI chips level, at that time they could not have met the demand, and it would have been chaos.
Curiously, He was going to be called «Chat with GPT-3.5»but the name was very long, so they shortened it to Chatgpt.
It seems the right decision, although Altman also recognizes that GPT-3.5 was not very good for chatgpt. Many people found errors, AI seemed silly, and they have not returned to it. Would it have had a better recognition if Chatgpt had premiered with GPT-4? Possibly yes, but the servers had hung every two by three.
At the end of the interview, the journalist explains that Grok is better in one thing: it treats users as adults. Sam Altman acknowledges: «I think we were very bad about that a while ago, but we have long improved.»
We have known some Chatgpt’s birth curiosities. Although it caused a revolution, its first version was very bad chatting. Also, when Grok launched, Sam Altman realized that they were far behind in the way of treating people as adults. They are lessons that Openai have served to improve.
Know How we work in NoticiasVE.
Tags: Artificial intelligence