According to OpenAI engineers, only 5 to 10 specialists are needed to retrain the GPT-4 model from scratch.
It is worth noting that hundreds of people participated in the creation of this model, released in March 2023, and almost all OpenAI resources were involved. In response to a question from the company’s CEO Sam Altman, the engineers answered that no more than 5-10 people are needed to retrain GPT-4 from scratch.
According to one of OpenAI’s employees, who managed the preliminary machine learning process for the company’s most powerful AI model to date, GPT-4.5, Alex Pineau, the retraining of the GPT-4o and GPT-4 models used much of the same material as the GPT-4.5 training. Daniel Selsam, a researcher at OpenAI who works on data efficiency and algorithms, agrees that rebuilding GPT-4 will now be much easier.
According to Sam Altman, the new model released in February GPT-4.5 was the first to give the impression of communicating with a thoughtful person. According to Pineau, GPT-4.5 was designed to be 10 times more intelligent than GPT-4. At the same time, Altman said that OpenAI no longer has computing power limitations for AI training, and this is something that the world has not yet fully realized.
Major tech companies are investing billions in AI infrastructure. This year, Microsoft, Amazon, Google, and Meta are expected to spend $320 billion in combined capital expenditures to expand their AI capabilities.
In March, OpenAI announced the closing of its largest-ever private technology funding round, including $30 billion from SoftBank and $10 billion from other investors, bringing the company’s valuation to $300 billion. The company said in a statement that the new capital will help OpenAI further expand its computing power.
Altman notes that at present, the OpenAI focuses on the effectiveness of the data on which AI learns GPT models are very efficient at processing information, but there is a limit to how deeply they can understand data.
He explained that with the growth of computing power, the amount of information is beginning to grow more slowly. To solve this problem, algorithmic innovations are needed to get more value from the same amount of information.
Don’t thank ChatGPT — OpenAI spends «tens of millions of dollars» on your courtesy
Source: Business Insider
Spelling error report
The following text will be sent to our editors: