The use of Language Models is becoming more and more popular in the industry, and OpenAI’s ChatGPT is no different. The computing power required to run ChatGPT is so extreme, according to semiconductor research firm SemiAnalysis, that it can cost OpenAI up to $700,000 a day. Microsoft is taking steps to reduce the cost by developing a secret Artificial Intelligence (AI) chip called Athena.
OpenAI is an AI research lab based in San Francisco, CA, with a vision to ensure that Artificial General Intelligence (AGI) will benefit humanity. Part of their research has gone into their GPT-3 language modelling system, popularly known as ChatGPT. The natural language processing (NLP) system values both accuracy and speed and can be used to solve complex tasks such as writing cover letters, generating lesson plans, and even redoing a dating profile.
While the cost to train ChatGPT’s large language models can exceed tens of millions of dollars, according to a phone call between Dylan Patel and Insider, the operational costs far exceed the training costs. Nick Walton, CEO of Latitude, commented on this during a conversation with CNBC where he said that running their AI could cost around $200,000 a month.
Microsoft is evidently seeking to reduce this cost with the development of a new AI chip. Reportedly, “Athena” is being developed by more than 300 Microsoft employees, and the chip is projected to be released by 2021. The objective of the chip is to help Microsoft and OpenAI reduce their costs of running the language models by drastically lowering their power usage.
Pavlo Gonchar is an analyst at SemiAnalysis whose estimates and knowledge of the expensive servers that ChatGPT requires has been integral to understanding the cost. In addition, Afzal Ahmad, another analyst at SemiAnalysis, has also enlightened people on the soaring running costs associated with language models.
AI21 Labs has developed another language processing software as an option, which reduces costs by 50% according to Nick Walton. AI21 Labs has been at the forefront of AI research, developing machine learning models and building AI stack components to further the advancement of AI technologies. This is a revolutionary step forwards in lowering the price of ChatGPT for startups and larger corporations alike.