ChatGPT users frustrated with the chatbot’s ignorance about recent events will now find it more informative and accommodating. At its first Dev Day conference for developers on Monday, ChatGPT creator OpenAI revealed that the knowledge cutoff dates for both GPT-3.5 and GPT-4.0 have been extended.
For GPT-3.5, the cutoff date has been increased to January 2022 from September 2021. And the cutoff date for the new version of GPT-4 dubbed GPT-4 Turbo is April 2023. These new timeframes mean you won’t bump into as many unanswered questions and apologies when asking the chatbot about the recent past. ChatGPT’s responses about relatively recent events should also be more accurate and precise.
The previous cutoff dates not only annoyed users but limited ChatGPT’s effectiveness as a research tool. The only way you could get real-time information was through the Browse with Bing option, a feature available just to ChatGPT Plus subscribers willing to shell out $20 a month. Even OpenAI CEO Sam Altman expressed frustration with the prior cutoff dates.
“We are just as annoyed as all of you, probably more, that GPT4’s knowledge ended in 2021,” Altman said at the Dev Day event. “We will try to never let it get that out of date again.”
To test the new cutoff dates, I ran ChatGPT in GPT-3.5 mode and asked a simple question: “What major events happened in November 2021?” The chatbot was able to provide a list of specific events that occurred during this month, including the 26th UN Climate Change Conference, ongoing efforts to combat the COVID-19 pandemic, two major infrastructure bills passed by the US Congress, and preparations to launch NASA’s James Webb Space Telescope.
For GPT-4.0, I had to tell ChatGPT not to use Browse with Bing to see if it could provide recent information without it. I asked it to tell me if Betty White were still alive without using Browse with Bing. In response, the chatbot said: “As of my last update in April 2023, Betty White passed away on December 31, 2021.”
I tried other questions focused on events during those months. In some cases, ChatGPT initially struggled to come up with the right information. I had to point out that its new cutoff dates should allow it to deliver the correct responses. After apologizing for its confusion, the chatbot was eventually able to provide accurate and updated results.