Subnostr

I've found ChatGPT isn't that great to help troubleshooting specific software, but is pretty awesome to use for troubleshooting code.

With specific software I think too many things are arbitrary or human dependent like names of things, whether the editing tool is called "cut" or "split," and that sort of thing. It often just makes up effects or tool names that don't even exist. Which makes sense because its a general language AI not a software specific AI (which I actually hope/think will be a really big thing in the coming years)

Whereas the syntax/naming conventions/commands in code is not app specific, but is universal across the language. Therefore it has far less of a problem pulling together related and specifically applicable chunks of code in response.

Reply to this note

Please Login to reply.

Discussion

radii 2y ago

Code is highly structured, llms do great with inferring that structure

Moss 2y ago

When natural language understanding has not been broken through, AI is only suitable for solving problems in a certain vertical field.

Now that natural language understanding has made a breakthrough, ChatGPT can understand the combined semantics between words, but there is a lack of data in vertical fields, so it can chat with you, but it cannot solve problems in a certain vertical field.

Now that Open AI opens up the API, it is hoped that all industries will come to help them train and instill vertical industry data. In this way, everyone can pay to help ChatGPT become stronger.

Now every industry with big data deserves to be changed by ChatGPT once.

#chatgpt