GPT-4 is here, and it’s already stunning early testers.
GPT-4 is the most powerful version yet of OpenAI’s large language model.
It is significantly better than the GPT-3 and GPT-3.5 models that power ChatGPT, providing outputs that display better reasoning, creative, and strategic capabilities than its predecessors. It also has the ability to use images, not just text, as inputs. And it can now handle up to 25,000 words at a time, so you can use GPT-4’s capabilities on long-form content.
(Before we go any further, the quickest way today to get access to GPT-4 is to sign up for ChatGPT Plus, OpenAI’s $20/month subscription plan for ChatGPT, which offers GPT-4 as a model option.)
In Episode 39 of the Marketing AI Show, Marketing AI Institute founder/CEO Paul Roetzer broke down exactly how GPT-4 changes the game. Here’s what you need to know.
“Reasoning is a really big deal,” says Roetzer.
The advanced reasoning capabilities of the tool have already been tested by having it take the Uniform Bar Exam. It scored in the 90th percentile. Compare that to ChatGPT, which only scored in the 10th percentile. GPT-4 also scored in the 99th percentile when tested on a national biology competition. ChatGPT only scored in the 31st percentile.
Overall, the outputs of GPT-4 versus previous models are proving to be much more logical, accurate, creative, and well-reasoned than before.
The model has the ability to understand images, a capability that isn’t yet available to general users. Roetzer guesses that will include the ability to process video at some point, too.
This ability to be “multimodal” makes GPT-4 much more powerful than text-only models. In one demo example, OpenAI showed off GPT-4’s ability to take a hand-drawn sketch of a webpage, understand it, and provide accurate code to build it.
“It’s moving much closer toward a general intelligence platform,” says Roetzer.
OpenAI says it has made GPT-4 significantly safer than previous models. It is, they claim:
“GPT-4 is 82% less likely to respond to requests for disallowed content and 40% more likely to produce factual responses than GPT-3.5 on our internal evaluations.”
However, OpenAI has not shared specific details on exactly how GPT-4 works. OpenAI will not say how large the model is or how it was trained. In turn, this makes it difficult for researchers to better understand the model and its risks.
To some, this approach seems dangerous and hypocritical. Instead of being the open research organization it started out as, OpenAI appears to be taking a highly closed approach to its work.
OpenAI co-founder Ilya Sutskever even told The Verge that “We were wrong” about sharing research in the past, given the competition in the market and the safety risks of extremely powerful AI falling into the wrong hands.
Agree or disagree with the approach, one thing is clear:
“They are very confident that having it open is dangerous to humanity,” says Roetzer.
You can get ahead of AI-driven disruption—and fast—with our Piloting AI for Marketers course series, a series of 17 on-demand courses designed as a step-by-step learning journey for marketers and business leaders to increase productivity and performance with artificial intelligence.
The course series contains 7+ hours of learning, dozens of AI use cases and vendors, a collection of templates, course quizzes, a final exam, and a Professional Certificate upon completion.
After taking Piloting AI for Marketers, you’ll: