ChatGPT’s web browser was too good, so its creators blocked it
It’s a rare day when a software developer blocks a feature for being too good, but that’s exactly what’s happened to OpenAI’s ChatGPT AI chatbot –its abilities to browse the web with Bing were simply too effective at dodging paywalls.
Recall that in March, OpenAI added support for the Bing browser to ChatGPT, specifically to give it knowledge of current events. Until then (and right now) the AI algorithm had only been trained up through the fall of 2021. If you asked it for the result of a recent sporting event, for example, ChatGPT would plead ignorance.
When Bing support was added, ChatGPT could then ferret out those answers, providing answers that were either current or just a day or so old. But OpenAI discovered that Bing was too good at its job — it was circumventing paywalls to provide the answers that users asked for. So OpenAI pulled the plug.
“We have learned that the ChatGPT Browse beta can occasionally display content in ways we don’t want,” the company said in a blog post. “For example, if a user specifically asks for a URL’s full text, it might inadvertently fulfill this request.”
Instead, asking ChatGPT today to summarize an article returns this response: “I’m sorry for the confusion, but as an AI developed by OpenAI, I don’t have the ability to browse the internet in real-time or access specific URL content. However, I can certainly provide general information on the topic based on my training data up until September 2021.”
Other, specialized plugins that take advantage of the internet, such as researching trips via Kayak, are still enabled, OpenAI said.
It’s worth noting that asking Bing Chat to circumvent a paywall doesn’t work. Instead, Bing notes that the article is “behind a paywall” and encourages you to sign up for the site. Google Bard may decline to provide the exact text, but repeatedly asking for the article’s content can eventually produce a detailed summary of what the article reports. It’s just another way in which Bing, Bard, and ChatGPT handle different queries.
Usually, OpenAI’s goals have been to provide what users want. In this case, however, noting that the Browse Beta can display content in “ways we don’t want” reflects the values that OpenAI is imposing upon the user. That’s good news for journalists who are using premium content to pay their bills and attract new readers, but disappointing for those who have seen the capabilities of AI chatbots become increasingly reined in. In this case, maybe OpenAI will find the happy medium that other sites use: simply “reblogging” the paid-for content?