What is GPT Best at Right Now?

Andy Greene
AI Land
Published in
2 min readMar 31, 2023
Photo by Roman Synkevych 🇺🇦 on Unsplash

Structured language aka programming languages

Languages that have really strict syntax. Logical flows. Only a few ways to express something.

This makes sense because the consistency of programming languages makes them easier to learn. In my experiments, GPT seems to get programming right about 80–90% of the time. It may not always be the most efficient code, but it does work.

Natural language has so many nuances (slang, incomplete thoughts, cultural references, horrible prompting). This makes it difficult to understand fully. I find about 50–70% of the time, it writes natural language correctly and with some originality.

When GPT 3.5 does natural language it is light years ahead of other models. Yet it gets stuck in loops. Its originality leaves something to be desired especially when it comes to advertising copy. It starts to be original and then it loops back to known language patterns. I think finetuning could solve this originality problem.

Like a small child, you have to keep checking in with GPT to make sure it stays on track. And it often takes prompting and re-prompting to get what you want.

Natural language will improve over time, but for now GPT is a game changer for programming languages and really cool for natural language.

Give it some time and it will be a game changer for natural language as well.

--

--

Andy Greene
AI Land
Editor for

Writing about marketing, big data, internet security and politics