Ask a chatbot for the weather in Boston, and it can now write the code to look it up. That is the example OpenAI used to show off “function calling,” a feature added to GPT-4 and GPT-3.5-turbo in an update covered by Tech Times on June 14, 2023.
The idea, per OpenAI’s blog post, is that programmers describe functions to the model. The model then generates code to run them. It is a tool for developers rather than a new consumer product, and Tech Times pointed out, fairly, that a quick search engine query gets you Boston’s weather just as easily.
Bigger memory, new pricing
GPT-3.5-turbo also received a larger “context window.” The model can now refer back to more of the preceding conversation, so its answers lose fewer details along the way.
The price changed in two directions. The advanced version of GPT-3.5-turbo costs twice as much as the standard one. The standard version got cheaper, with its price cut by 25 percent, according to Mashable.
Who already builds on OpenAI
Search Engine Journal, cited by Tech Times, listed a string of companies that use OpenAI’s tools in their products.
Microsoft and OpenAI have worked together to give developers AI models and to improve Bing and Office. Snapchat’s chatbot My AI runs on OpenAI GPT models, and its latest update lets it send and interpret image snaps. Salesforce unveiled Einstein GPT, billed as the first generative AI customer relationship management product.
Morgan Stanley partnered with OpenAI for access to GPT-4, making it one of the select few wealth management firms with the technology. HubSpot built ChatSpot.ai on GPT-4. Stripe uses GPT for customer insights and fraud detection, and GetResponse added an email generator powered by it. Instacart’s shopping chatbot relies on OpenAI too.
GitHub Copilot, the code completion tool, integrated OpenAI Codex. That integration has come with a legal dispute over copyright.
