OpenAI, the company behind GPT-3 and ChatGPT, today announced the release of GPT-4. This is a major upgrade to the amount of textual and image data that GPT relies on for its (mostly) uncannily accurate and verbose responses. OpenAI says that “GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5.”
The dataset is clearly larger than the earlier versions, though OpenAI is cagey as to the current size. The last version appeared to know little or nothing beyond early 2020; GPT-4 is trained with data up to September 2021.
GPT-4 is a so-called “multimodal large language model,” meaning it responds to both text and images. In the product demo, a picture of food in a fridge was used to generate recipes for the range of leftovers on display.
OpenAI readily admits that GPT-4 has the same limitations as previous GPT models and is not fully reliable—it “hallucinates” facts with great confidence. But the extensive data published on the product web page shows that the new version performs far better than its predecessors. OpenAI claims that GPT-4 is 82% less likely to respond to requests for content that isn’t currently allowed, and 60% less likely to make stuff up.
“It’s not perfect,” Greg Brockman, president and co-founder of OpenAI, said during the product announcement live-streamed on YouTube. “But neither are you.”
The publishing community has struggled with a range of concerns around GPT and ChatGPT, primarily surrounding the ease with which low-quality text can be generated for short stories and full-length books. Some publishers are reporting a significant increase in low-quality submissions. Amazon is hosting short AI-generated books. At the same time, a host of copyright issues perplex the creative industries, from images to music to text, leading to, for example, proposed restrictions from the Authors Guild.
But there are optimists. Ken Brooks, of Treadwell Media Group, says that “while the introduction of GPT-4 will further blur the lines of human and AI-generated content, the biggest short-term impact will be behind the scenes in translation, personalization of content and learning materials, enhanced marketing, curation and summarization.” But he expects this still to remain “in the hands of competent authors, editors and production staff.”
Brooks also asked GPT-4 to “generate a zippy quote on the importance of the release of GPT-4 to publishers.” The response: “With GPT-4’s arrival, the publishing world witnesses a paradigm shift, unlocking the gates to a universe of creativity, unparalleled efficiency, and limitless literary possibilities—a true renaissance in the art of the written word!”