After launching Silicon Valley’s chatbot sensation ChatGPT, maker OpenAI is expected to announce product enhancements that will make its AI models cheaper, with more functionality. The company’s first-ever developer conference begins Monday, and CEO Sam Altman has teased attendees with the promise of “some great new stuff.”
The updates are intended to address concerns that developers have had about the cost of using OpenAI’s generative AI technology, which is a form of AI that generates completely new content such as text, images, or software code. They will also help developers bring these advanced features to more applications.
According to a Reuters news report, the first major update is a 75% price reduction for GPT 3.5 Turbo, the model that powers ChatGPT and other generative AI services. This is significant because the underlying models are costly to train. OpenAI says the price reduction is made possible by improved efficiency in the model used to create the responses to user queries. The more efficient system makes it less costly to train the model and pass those savings on to developers.
In addition, the company plans to introduce vision capabilities that enable the model to analyze and describe images. This opens up a variety of new applications, such as image search. It’s also a way to allow users to interact with machines more visually, which could be especially useful for visually impaired people.
Another change involves how developers can fine-tune GPT-4, the most advanced of the company’s models. Currently, the model is set by default for specific context windows, but developers can use a tool to create a custom configuration and assign it to the model. This will let developers customize the model’s behavior and better tailor it to their particular application.
Over a year since the ChatGPT craze started, the company’s generative AI has already been used to create various products, from writing assistants to doordash apps. It’s also been used by Microsoft to power trippy results for its Bing search engine.
Despite the popularity of these technologies, there are still many critics who fear that they can be used to manipulate users or spread false information. OpenAI has addressed some of those concerns by implementing a feature that prevents the model from generating content it thinks is inappropriate, such as racist or conspiratorial material. The company has also worked to broaden the views represented in its model by hiring more data labelers from diverse backgrounds.
In a move that signals its ambition to expand beyond a consumer sensation to becoming a provider of a hit developer platform, the company is introducing new tools that will make it easier for developers to add its generative AI models to their apps. By making its models more affordable, enabling more flexible and powerful customizations, and adding vision capabilities, the company hopes to make its technology indispensable for outside developers.