Launched just a few months ago, ChatGPT has changed many people’s perceptions of what AI can do. The AI at that time was based on GPT-3.5 from OpenAI, which was also integrated into Microsoft Bing, Skype and Edge. Now the company has confirmed that it has switched to the new and more powerful GPT-4 model. In fact, the new software has been running for some time. Anyone who has tested the Bing preview has been using GPT-4 for five weeks.
GPT-4 is a “multimodal” model, meaning you can combine your text requests with images. Here’s an example of GPT-4 as it explains a Reddit joke: What’s funny about a Lightning cable in the form of a VGA cable. However, the answer contains only text, you cannot generate images like with Stable Diffusion or MidJourney.
The new model has also become smarter, which the OpenAI team tested with exams from 2022 and 2023. GPT-3.5 passed the bar exam, but was classified in the bottom 10%. GPT-4, on the other hand, scored in the top 10%. The justice system may not be ready for robo-lawyers, but they appear to be on the horizon. On the LSAT exam, GPT-4 was in the 88th percentile, v3.5 was in the 40th percentile. And on SAT Math, GPT-4 was in the 89th percentile, GPT-3.5 in the 70th percentile. See the OpenAI announcement for more audit results.
The most important innovation in version 4 is “steerability”. So far, ChatGPT has behaved like a digital assistant subject to a few rules. You could get the AI to reveal those rules. However, Microsoft and OpenAI have worked to hide such rules to prevent so-called “jailbreaking”. With GPT-4, on the other hand, companies can actively control the style and behavior of the AI.
It’s important to note that GPT-4 still has limitations, especially when it comes to facts. Like its predecessor, the model can make up things, what OpenAI calls “hallucinations”. But the new version is much better at sticking to the facts and not making any logical mistakes. However, GPT-3 was released in mid-2020. GPT-3.5 came in early 2022 and provided the basis for ChatGPT.
