Broken motivator
Winkletter • 1 Dec 2023 •
People have been noticing, myself included, that ChatGPT is getting lazy. Coders especially have noticed where the chatbot is leaving code unfinished with a note to the developer to finish the code.
I feel like the problem might be related to the token length. There is talk that the token length of responses was cut down after the OpenAI conference to 850 tokens. If you ask the chatbot to write a program that requires a lot of code it might need to wrap up the response, and the easiest way of doing that is to push back on the user to finish the code.
Nah, I don’t feel like it. It’s too cold.
With the writing exercises I’ve been generating, I’ve seen where ChatGPT isn’t finishing the example at the end of the exercise. Pretty consistently it is only providing an example of the first four instructions. I haven’t minded this since I end up rewriting much of the example myself, anyway.
But I think it shows yet again why it’s important to have something running locally or where you have more control over the tool.
Comments
I’m not a power user, so I haven’t noticed this. I knew it was too good to last.
I’m still using it myself. I can change my prompts to expect shorter replies. But I do hope to get some experience soon working with the models directly.
@Winkletter Have you seen the prompt add-on to tip ChatGPT with money? They say it will encourage longer prompts haha. WTH. Another one I seen is say you got some disability or unable to type for some reason and it will give you the longer replies
Interesting. I wouldn’t want our relationship to be built on lies, so I’ll just keep treating ChatGPT with respect and admiration.
Seeing this complaint too on Twitter. Makes me wonder if I should subscribe to Plus… been waiting for them to reopen the gates, but this is making me doubt