Chatting with the ChatGPT chatbot, developed by OpenAI, could cost at least $700,000 a day, according to Dylan Patel, chief analyst at SemiAnalysis, a firm that specializes in semiconductor industry analysis, speaking to The Information.
This cost is primarily related to the use of high-performance servers running the software. It is also probably revised down, because Dylan Patel specifies that this cost is estimated in GPT-3, the version ChatGPT was based on prior to the March release of GPT-4.
exorbitant costs
Containing Forbes, the analyst noted that, in addition to the budget required to train the chatbot in various language models, operational expenses “far exceed training costs when ChatGPT is deployed on such a scale. In fact, generating text with ChatGPT is more expensive than training it weekly.”
And the scale at which the bot is deployed is quite compelling: the company shows numbers to back up its success, with more than 186 million unique users in the month of March. Ultimately, all costs combined, ChatGPT would cost millions of dollars per day.
Encouraged by the success of its commitment to OpenAI, Microsoft would have another card to play: manufacture its own chip. To reduce the exorbitant costs of artificial intelligence, the company is secretly working on manufacturing this new chip called “Athena”, according to The Information.
Its purpose would be to operate and train its artificial intelligences, and it would be in the works from 2019. Its use, only internally, could start as soon as next year, according to the media.
Source: BFM TV
