All but a handful of artificial intelligence startups typically fall into one of two camps.
The first group uses a single large-language model, typically OpenAI’s GPT-4, to power their applications. These startups, which have often raised millions in venture capital, usually say, “If you’ve got the cash, why not use the top-of-the-line model?”
But as the number of commercially available LLMs continues to rise (there are now at least 300,000 to choose from on Hugging Face alone!) as open-source models proliferate, I’ve noticed more companies using those models in addition to proprietary ones from OpenAI and its ilk. This strategy can help the startups cut costs by routing simpler queries to cheaper, less powerful models. It can also help them get around OpenAI’s rate limits, which cap how much a company can access GPT-4 through an application programming interface.
0 Commentaires