It can now pass a bar exam in the top 10% previously GPT-3 came in the bottom 10%.
In casual conversation the difference can be subtle but in the more complex task GPT-4 is more reliable, creative, and can handle more nuanced instructions.
GPT-4 can also accept input of text and images.
It also still ‘hallucinates’ facts and makes reasoning errors. It can be confidently wrong in its predictions and will not double check them.
ChatGPT Plus subscribers will get access with a usage cap, which will be adjusted depending on usage and demand.
Grief. It was only scheduled to be ready in July/August/September. It’s learning more, faster. It’s starting to learn from visuals combined with text as well. And in foreign languages now too.
Still, there’s hope that our mate Simpiwe van der Hoosen might still live a little longer…
Despite its capabilities, GPT-4 has similar limitations as earlier GPT models. Most importantly, it still is not fully reliable (it “hallucinates” facts and makes reasoning errors).
Compared to the earlier versions, it’s brighter than ever:
I wonder if they’ll need to build in a “fail-safe” at any point so that the algorithms intentionally put in invalid information, or make small errors on the output to avoid the systems becoming too “smart” or people being too stupid.
