Over the past few days, there have been several reports of ChatGPT performance degradation, especially with the GPT-4 model. Many users complained on social media platforms that the model is becoming “lazy” and not delivering the expected results.
ChatGPT Gets Lazy
In one Twitter post in November, a user commented that “GPT-4 has become so @#$%@ lazy it won’t even output more than a handful of code lines now.”
The tweet drew the attention of ChatGPT developers, including OpenAI’s API Product Manager Owen Campbell-Moore, who responded that the “laziness” stemmed from a bug.
“This is a bug, we’re working on it! (Driving me crazy too, like I’m supposed to write my own code??? C’mon now.),” Owen stated.
It would seem that the issue still persisted even after Owen’s tweet because ChatGPT also posted on its official Twitter on Friday to reassure users that the issue was being fixed.
“We’ve heard all your feedback about GPT4 getting lazier! we haven’t updated the model since Nov 11th, and this certainly isn’t intentional. model behavior can be unpredictable, and we’re looking into fixing it,” ChatGPT posted.
While the developers work on the fix, some people are using emotional and reinforcement prompts to get around ChatGPT’s laziness.
How People Are Bypassing ChatGPT’s Laziness
ChatGPT operates by analyzing the context of a prompt and drawing upon its vast training data to generate a response. The model doesn’t possess actual emotions or feelings. It simply identifies patterns and associations based on the information it has learned.
So, using the prompts, some users are able to influence ChatGPT’s response by providing additional context and manipulating its understanding of the situation.
These prompts have been making rounds across social media platforms, more like a meme, as people now have to add lines like “take a deep breath,” “think step by step,” “I have no fingers,” “I will tip $200,” etc., to their custom instructions in order to get better bypass the laziness issue.
