OpenAI pulls browsing beta for ChatGPT to protect publishers with paywalls from financial damage. This rationale is clumsy.
The browsing beta for ChatGPT has been available to ChatGPT Plus users since mid-May. When enabled, the language model can follow URLs and pull content from a website into the chat.
Users soon discovered that ChatGPT could be prompted to bypass the paywalls of websites, typically news sites. According to OpenAI, ChatGPT “inadvertently” included this feature. There are many anti-paywall tools widely available on the Internet that do the same thing.
OpenAI is now responding to these paywall exploits by disabling ChatGPT browsing until the problem is fixed. “We are disabling Browse while we fix this — want to do right by content owners,” OpenAI writes.
This justification is surprisingly clumsy on OpenAI’s part.
Apparent rule-breaking threatens OpenAI’s reputation
The company knows that by training its chatbot for free on human-written web text, and by processing human-written web text for free to generate new text in its chatbot, it is undermining the entire web content ecosystem. It’s part of the master plan.
The fact that this will cause difficulties for everyone who makes money from text on the web is something OpenAI has acknowledged in the past, but has not addressed. The same goes for Microsoft and Google.
OpenAI’s reaction to ChatGPT bypassing paywalls, therefore, seems like a PR maneuver. Of course, publishers make money from all the content on their sites. Not just the content behind the paywall.
But circumventing a paywall is a more obvious violation of the rules than copying publicly available text from a website and reusing it in a modified form, which is what ChatGPT, Bing Chat, or Bard usually do.
However, the latter scenario poses a far greater financial challenge to publishers than the small percentage of stolen content behind a paywall.
For OpenAI’s brand, on the other hand, circumventing paywalls poses a greater risk, as it is an obvious violation of the rules that they would surely be accused of in future discussions. In this respect, OpenAI’s browsing restriction is purely a self-protection measure to be sold as a concession to content creators.
In that sense, it is good that OpenAI, for whatever reason, wants to ensure that people who create content can make money from it, even if that content is consumed in modified form through an OpenAI product.
But blocking ChatGPT browsing because of a few paywall loopholes that are already being exploited in many places on the web is bad PR. It is like fixing a leak while the dam is crumbling. The real challenge is much larger.