"OpenAI on Thursday released a smaller and cheaper version of the technology that powers ChatGPT, as the company known for cutting-edge artificial-intelligence research tries to become a more successful business.
The new version of its AI model, called GPT-4o mini, is 60% cheaper to use than the model that powered ChatGPT until recently, called GPT-3.5 turbo, the company said, and scores higher on tests it uses to measure intelligence and effectiveness.
OpenAI is the best-known brand in the booming space of generative artificial intelligence, so its embrace of efficiency is a notable indicator of a broader turn for the industry. Previously, OpenAI and competitors like Google and Microsoft focused primarily on developing the biggest and most powerful AI models. Now they are trying to balance those pushes with releasing small models that can be more profitable for them and more useful for business clients that only need AI tech to do certain tasks well.
Google, and startups Anthropic, Mistral and Cohere have also released smaller models this year. Microsoft has played up a family of small models named Phi it said was one-hundredth the size of the model behind ChatGPT at the time.
Many generative AI startups, as well as new entrant Apple, also are focusing on small models that are cheaper to build because they are typically trained on smaller batches of data. They also require less computing power to operate and can in some cases work directly off hardware like a phone.
OpenAI said GPT-4o mini can currently interpret text and image inputs and will later add the capability to scan and output audio and video. The technology was rolled out to people who use the free and premium versions of ChatGPT starting Thursday. The company said it would be available to corporate customers next week." [1]
1. ChatGPT Maker Goes Smaller, Cheaper With New AI Model. Perry, Alex. Wall Street Journal, Eastern edition; New York, N.Y.. 19 July 2024: B.4.
Komentarų nėra:
Rašyti komentarą