When I logged into OpenAI’s website last year to continue testing the new version of ChatGPT powered by GPT-4, the chatbot didn’t try to sabotage my relationship, write my emails, or unleash my creativity—it simply didn’t work. Demand was high that day, and the company was experiencing occasional outages. Greg Brockman, an OpenAI cofounder and president, was upfront about the model’s imperfection in a livestream around the same time. He also reminded listeners that they were not without blemishes themselves. Generative AI remains a focal point for many Silicon Valley developers after OpenAI’s transformational release of ChatGPT in 2022. The chatbot uses extensive data scraped from the internet and elsewhere to produce predictive responses to human prompts. It was previously powered by the GPT-3.5 language model. While that version remains online, an algorithm called GPT-4 is also available with a $20 monthly subscription to ChatGPT Plus. If you’re considering that subscription, here’s what you should know before signing up, with examples of how outputs from the two chatbots differ. The core service you pay for with ChatGPT Plus is access to GPT-4. You’re not buying an unlimited number of prompts, though. Subscribers are currently limited to inputting 40 prompts to GPT-4 every three hours. After hitting the prompt limit, users can always switch to the GPT-3.5 version. (In addition, you can experiment with a free version of GPT-4 by signing up for Microsoft’s Copilot chatbot.) A ChatGPT Plus subscription also unlocks OpenAI’s newest software features. For example, Dall-E 3 is available inside the bot to generate images for you via text prompts. Subscribers can also use ChatGPT’s Bing integration to browse the web in real time for information that may be relevant to their prompts. Another core aspect of ChatGPT Plus is the GPT Store. Here’s where you can access versions of OpenAI’s bot that have been customized by the community with additional data and parameters for more specific uses, like coding or writing help. You can even try out a unique bot that’s based on my writing for WIRED. Many aspects of GPT-4 remain opaque. OpenAI has not shared many details about GPT-4 with the public, like the model’s size or specifics about its training data. With that in mind, are answers from the GPT-3.5 and GPT-4 models noticeably different? OpenAI’s website reads, “GPT-4 excels at tasks that require advanced reasoning, complex instruction understanding, and more creativity.” It seems like the new model performs well in standardized situations, but what if we put it to the test? Below are the two chatbots’ initial, unedited responses to three prompts we crafted specifically for that purpose last year. Curious about other chatbot subscriptions? Check out our head-to-head comparison of OpenAI’s ChatGPT Plus and Google’s Gemini Advanced, which also costs $20 a month. GPT-3.5: Knock knock. Who’s there?
