GPT upgrade: OpenAI's latest model is a big step up

Not yet a subscriber? Sign up free below.

The world of artificial intelligence just got a lot more exciting, or scary, depending on your perspective, as the tech firm behind the wildly-popular and surprisingly-addictive ChatGPT unleashed its new model.

OpenAI’s latest creation, GPT-4, is a smarter, more powerful version of the tech behind the viral chatbot — and on a series of benchmark tests, the engine’s ability to outperform humans is staggering.


While GPT-4 is locked behind OpenAI’s paywall, Microsoft revealed that it’s been powering Bing’s chatbot for 6 weeks now. Even though that test of the language model’s capabilities hasn’t always run smoothly, the new version has outscored its predecessor on nearly every academic and professional exam.

GPT-4, for example, would apparently beat a staggering 90% of lawyers attempting to pass the bar, compared to GPT-3.5 which would have been in the bottom 10%. Indeed, in 12 of the 15 AP exams the pair took, GPT-4 scored ahead of 50% students. The model still has some revision to do in the English department, however, showing no improvement in language or literature.

Aside from acing exams, GPT-4 is also able to accept images as inputs — a key factor in one of its first implementations where the model will be used to power a “virtual volunteer” for those with visual impairments.

We're a fully bootstrapped, independent media company. If you enjoyed this data story, sharing it with your friends helps us out a lot.
Not yet a subscriber? Sign up free below.


Stories from this newsletter

GPT upgrade: OpenAI's latest model is a big step up
Willow windfall: A controversial Alaskan oil project has got the green light
Global convenience: The rise of 7-Eleven