The latest version of ChatGPT, OpenAI’s groundbreaking generative AI chatbot, now has access to real-time information from the internet at large, curing what had been seen as one of its key flaws.
Thanks to an integration with Microsoft’s Bing search engine – the company is a major investor in ChatGPT maker OpenAI – the AI chatbot is now able to look up the latest available information on any topic. Previously, ChatGPT's answers were based merely on its large language model (LLM) training data, which only ran through September of 2021.
The browsing feature was originally launched in June 2023, but disabled shortly thereafter, due to what OpenAI called a habit of “occasionally display[ing] content in ways we don’t want,” which reportedly included skipping through paywalls to access content it shouldn’t have. Browsing rolled back out to beta customers of OpenAI Plus (the company’s paid subscription service, which includes access to the more-advanced version 4 of the AI) late last month, and is now out of beta and fully available to Plus customers.
“Plus and Enterprise users no longer need to switch the beta toggle to use browse, and are able to choose ‘browse with Bing’ from the GPT-4 model selector,” OpenAI said in release notes published this week.
Users will still need a Plus subscription to access the new functionality, and can reach it from within the OpenAI app or Bing browser. The company also added direct access to the DALL-E 3 image generation system from within the ChatGPT client this week, which also requires a paid subscription. That feature is only available in beta, however, meaning it’s not yet available to every user.
“From a simple sentence to a detailed paragraph, ask ChatGPT what you want to see and it will translate your ideas into exceptionally accurate images,” OpenAI’s release notes said.
Other chatbots already boast some access to the internet, although the precise uses of internet connectivity vary. Perplexity, an open source chatbot based on ChatGPT v3, has had the type of internet access now implemented in OpenAI’s proprietary version for some time, while Meta’s in-house LLM is able to use Bing for image generation from text prompts. (Meta’s focus has been different than OpenAI’s in several ways, centering more on providing an array of specific-use chatbots rather than one overall general intelligence, although the Facebook parent company does have Llama LLM to fill that role.)