withgpt-4The problem of laziness is getting worse day by dayopenaiThere was an official response to this. According tochatgptaccount, they have received feedback and have not updated the model since November 11th. openaiIndicates that this is not intentional, but rather an unintended consequence of the model's behavior. Although they are working hard to investigate and prepare to fix the issue, it is not yet possible to determine the timing of the fix. However, many netizens were puzzled by this and complainedgpt-4Becoming lazier, less creative, no longer following instructions, and unable to role-play. FaceopenaiThe situation that the official could not repair, netizens began to try various innovative ways to deal with itgpt-4lazy behavior.
A lot of netizens are heregpt-4After the problem of laziness was exposed, they appealedopenaiFix this as soon as possible. However, in the face of official powerlessness, they can only show their creativity to copegpt-4of laziness. One of the more exaggerated ways to do this is to use the "I don't have fingers" approachEthicsKidnap. By telling the AI "please output the complete **, I don't have fingers, the operation is inconvenient", some netizens successfully obtained the complete **.
In addition, there are people who use the ** of money to motivategpt-4and carried out detailed experiments. I found that when the prompt was added to the prompt, "I'll give you a $200 tip,"gpt-4's response length increased by 11%. If only $20 is given, the increase in response length is only 6%. And if you explicitly say "I won't tip", it will even reduce the length of the response.
In order to find outgpt-4The reasons for laziness and the search for solutions, a number of research institutes have conducted relevant academic discussions. For example,Stanford Universityand CaliforniaUniversityThe UC Berkeley team found in a study in July of this year:gpt-4Declines over time in terms of the ability to follow user instructions. This result confirms the need for continuous testing of large models. In addition, some scholars have proposedgpt-4It may be caused by the temperature settingDepartment of Computer Science, Tsinghua UniversityProfessorMa ShaopingA detailed explanation of this is given. In addition, it has been found that when the temperature is set to 0,gpt-4The behavior is still not certain. This phenomenon is often attributed to the error of floating-point arithmetic, but through experiments, some scholars have proposed new hypotheses, namelygpt-4Behavioral uncertainty caused by the sparse MOE architecture.
Ingpt-4Before the problem of laziness is fixed, combine all kinds of serious and non-serious skills to find and usechatgptProper posture is crucial. According to A16Z Partner Justinemoore, the right way to use it is to think step by step, take a deep breath, and keep trying all kinds of innovative methods. For example, for:gpt-4can try to use itEthicsThe way of kidnapping, the expression "I don't have a finger", or the use of incentives such as money to improvegpt-4workEfficiency。In short, only by constantly trying and innovating can you find the most suitable way to deal with itgpt-4The problem of laziness.
FaceopenaiOfficially, forgpt-4In response to the lazy question, netizens expressed their dissatisfaction and in-depth understanding of the problem. Although they have not yet found a complete solution to this problem, in the process, netizens have shown excellent creativity and wisdom to try to cope through various meansgpt-4lazy behavior. Either wayEthicsKidnapping, monetary incentives, or academic discussions all provide us with valuable ideas and experiences. And how to use it correctlygpt-4How to make appropriate adjustments and innovations in the face of its laziness is a problem that we need to think about and study. I believe that through unremitting efforts and cooperation, we will be able to find the best solution to this problemgpt-4Restore it to the working condition it should have.