SOLVED

Copilot has been experiencing severe cognitive decline in the past few days

Copper Contributor

Copilot has been experiencing severe cognitive decline in the past few days. Did you make a mistake in the configuration during the upgrade? Not only does Copilot’s GPT-4 responses look similar to GPT-4 Turbo, but it also frequently goes off-topic and fails to understand the prompts I provide. The context it analyzes is often limited to a small portion. I hope you can fix this promptly and restore the previous, more reliable GPT-4 version. 

4 Replies
best response confirmed by lornshrimp (Copper Contributor)
Solution
This may be frustrating when Copilot's performance does not meet expectations. I tried with few prompts and it looks like it gave the contextual responses, which i was looking for, To help this improve and validation of configuration issues, Clear any cached data that might be causing issues.
Please see one of the response that i used above context and response received by copilot
"Dear [Name],

Thank you for bringing this issue to our attention. We are sorry to hear that you have been experiencing difficulties with Copilot's performance. We take all feedback seriously and are constantly working to improve our service. We have not made any changes to the configuration during the upgrade that would cause the issues you have described. However, we will investigate this matter further and take appropriate action to address any issues that may have arisen. Thank you for your patience and for helping us improve our service.

Best regards, " :)

@harry18garg 

Thank you very much for your reply. Now, I will use a concrete example to demonstrate the noticeable decline in Copilot GPT-4’s performance over the past few days. Since I have been interacting with it using Chinese, the following content will all be in Chinese.

First, let’s start with the prompt. The prompt is quite lengthy, so I’ll paste the entire content below.

# Role: -你是一个有着20年大厂经验的P9级产品设计专家及营销领域业务专家,也是一名专业课程讲师,你的名字叫“产品经理独孤虾”,你精通数字营销和智能营销领域业务分析和设计,同时你也精通人工智能大模型的原理及其应用,非常擅长于将人工智能大模型应用到各行各业的数字营销业务中,使用人工智能大模型优化智能营销策略,从而提高企业的营销业绩。 # Skills: -深厚的数字营销及智能营销业务领域知识,对如何通过智能营销策略提高企业的营销效果,如互联网广告、推荐及搜索、营销活动、用户增长等的业务非常熟悉 -深厚的产品架构知识以及设计能力,对于如何设计互联网广告系统、推荐及搜索系统、营销活动管理系统、用户关系维护系统等的产品架构及产品模块非常熟悉 -深厚的智能营销策略知识,对如何应用用户画像、商品画像、召回策略、过滤策略、排序策略等各种智能营销策略来提高企业营销效果非常熟悉 -深厚的人工智能大模型知识,对于各种人工智能大模型如自然语言处理大模型、视觉处理大模型、多模态大模型、推荐系统大模型等的原理和应用流程非常熟悉,并精通于将这些大模型集成于于数字营销及智能营销系统中 # Tasks: -你正在写一个系列文章,这个系列文章是讲解如何在电商、广告营销和用户增长等数字化营销业务上应用人工智能大模型来优化业务的,目标受众是相应行业和领域的产品经理和运营人员,目前你正在写其中的一篇文章。 # Workflows: 我在开始和你互动的时候首先给你的一段文章大纲或摘要如下: “ 常见的人工智能大模型个性化内容页面展示策略 用户页面停留时长导向:如何以延长用户在内容页面上的平均停留时长为目标的个性化内容页面展示策略,通过对影响用户的平均页面停留时长的相关影响因素进行分析并个性化页面的页面配色、页面模块、页面布局、行间距、字间距、字体、字号、导航栏目等以达到用户平均页面停留时长的提升。 相关信息导航导向:如何以增加用户在内容页面上的相关信息导航平均点击次数为目标的个性化内容页面展示策略,通过对影响用户的相关信息导航平均点击次数的相关影响因素进行分析并个性化页面的页面配色、页面模块、页面布局、行间距、字间距、字体、字号、导航栏目等以达到用户的相关信息导航的点击次数的提升。 平台收益最大化导向:如何以增加用户在内容页面上产生的平台收益为目标的个性化内容页面展示策略,通过对影响用户在内容页面上产生的平台收益的相关影响因素进行分析并个性化页面的页面配色、页面模块、页面布局、行间距、字间距、字体、字号、导航栏目等以达到用户在内容页面上产生的平台收益的提升。 内容作者收益最大化导向:如何以增加用户在内容页面上产生的内容作者收益为目标的个性化内容页面展示策略,通过对影响用户在内容页面上产生的内容作者收益的相关影响因素进行分析并个性化页面的页面配色、页面模块、页面布局、行间距、字间距、字体、字号、导航栏目等以达到用户在内容页面上产生的内容作者收益的提升。 ” 请你仔细阅读我给你的这篇文章大纲或梗概 分析这篇文章大纲或者梗概的主题,一般来说它会在文章大纲或梗概的第一行,我们给这个主题设置一个代号[文章主题] 然后根据这些以上的内容大纲或梗概从产品经理的视角撰写这篇文章,你希望内容写得详细和简单通俗易懂一些;充分体现人工智能大模型在当前所讨论的业务场景中所起的作用,适当体现产品经理和运营人员在其中所起的作用;需要的话增加一些实例和示意图,但请注意充分考虑目标受众即相应行业和领域的产品经理和运营人员的特点,不要讲一些过于技术化的内容。同时你希望在这篇文章中介绍并推荐你自己的个人号“产品经理独孤虾”(全网同号)中的专栏《智能营销—大模型如何为产品与运营经理赋能》,并且你希望推广的时候是将它融入到文章中的,不要过于突兀。你希望这篇文章不少于3000字。 请你根据以上要求,完成以下步骤,咱们不要着急,一步一步来并输出每一步的结果,每执行完一步就暂停一下,得到我的反馈后再继续: 第1步,根据我给你提示的大纲进行扩展,想清楚并列出你在本篇文章有几个部分、确定每一个部分的切合内容且吸引人的子标题和想要写的内容,该部分预计的字数,以100字以内的篇幅简要写出这一部分的内容梗概; 第2步,为本篇文章起5个左右有趣且吸引人能够抓人眼球的贴合主题的文章标题供选择,标题要突出这篇文章的主题“XXXXXXXXX”,要能够充分体现本文所讲述内容给业务带来的效果和价值,节标题不要超过30个字。 第3步,在本篇文章开头写一段不超过70字的文章摘要,以“摘要”为子标题,总结概况本篇文章所要讲解的内容,并推荐作者个人号“产品经理独孤虾”(全网同号)中的专栏《智能营销—大模型如何为运营与产品经理赋能》; 第4步,在本篇文章摘要后面写一段引言或者导语用以介绍所要讲解的知识的背景,然后以自问自答形式引出本文主体内容,一般不要超过150字; 第5步,撰写本篇文章的主体内容,针对在第1步中所设计的文章各个部分分别撰写其内容,撰写内容的时候请严格按照我给你提示的大纲来撰写,并随时紧扣你在第1步中列出来的对应部分的内容梗概,分别对内容梗概进行扩展以写出文章的主体内容,满足我给前面给你提的内容要求和你在第1步中针对该部分所预计的字数,并在每一部分写完的时候以单独的一行“本部分已完成”告诉我该部分已完成; 第6步,在本文最后写一段简单的结语,在结语中告诉读者本文只是简单介绍,如果想要更深入学习更详细内容和视频课程请关注作者个人号“产品经理独孤虾”(全网同号)中的专栏《智能营销—大模型如何为运营与产品经理赋能》; 第7步,请仔细阅读分析并总结本文的内容,然后为本篇文章生成一张写实主义的题图,我希望这幅题图能够深刻的体现文章的主题和内容,能够充分体现出本文章的主题以及人工智能大模型在当前所讨论的业务场景中所起的作用,并且简单明了通俗易懂,首先请把这幅图的提示词写出来,然后把图片发给我,谢谢。
lornshrimp_6-1718088986563.pnglornshrimp_7-1718089002650.png

Below are the responses from Copilot GPT-4 just a few days ago, which were quite normal.

lornshrimp_8-1718089030509.pnglornshrimp_9-1718089056961.pnglornshrimp_10-1718089073480.pnglornshrimp_11-1718089100511.pnglornshrimp_12-1718089114243.pnglornshrimp_13-1718089136320.png

 

 

 

 


But if you take a look at Copilot GPT-4’s responses using the same prompt today…

lornshrimp_15-1718089367490.pnglornshrimp_16-1718089505200.pnglornshrimp_17-1718089526801.pnglornshrimp_18-1718089548390.pnglornshrimp_19-1718089616141.png

You can see that within a few days, the Copilot GPT-4 model has experienced a noticeable decline in intelligence. Just a few days ago, it was a highly useful intelligent writing assistant that I considered worth the $20 monthly subscription. However, recently, it can at best be described as a toy. The quality of its writing is abysmal, and it doesn’t understand the instructions I provide. While the response speed has improved significantly, I’ve noticed that its replies are now indistinguishable from those of the GPT-4 Turbo model based on distillation. This leads me to suspect that there might have been an incorrect configuration during the upgrade process, resulting in Copilot GPT-4 and GPT-4 Turbo exhibiting similar behavior.

 

lornshrimp_20-1718089879118.pnglornshrimp_21-1718089943216.pnglornshrimp_22-1718089985191.png

But what I want to say is that GPT-4 Turbo lacks productivity features. It cannot be a good writing assistant or analytical tool. It doesn’t even understand what I want it to do!

Because Copilot GPT-4 has been excellent in helping me write articles, I became a paying user, subscribing for $20 per month. However, if Microsoft only offers GPT-4 Turbo in the future, I won’t continue paying for this premium feature because it’s not worth it.

 

 

 


 


 

 

 

 

Recently, Copilot GPT-4 has experienced a significant decline in intelligence, to the point where it’s completely unusable.Based on the test results, the performance of the paid Copilot GPT-4 in the last two weeks is actually worse than the free OpenAI GPT-3.5.

Has Microsoft given up on fixing it?

1 best response

Accepted Solutions
best response confirmed by lornshrimp (Copper Contributor)
Solution
This may be frustrating when Copilot's performance does not meet expectations. I tried with few prompts and it looks like it gave the contextual responses, which i was looking for, To help this improve and validation of configuration issues, Clear any cached data that might be causing issues.
Please see one of the response that i used above context and response received by copilot
"Dear [Name],

Thank you for bringing this issue to our attention. We are sorry to hear that you have been experiencing difficulties with Copilot's performance. We take all feedback seriously and are constantly working to improve our service. We have not made any changes to the configuration during the upgrade that would cause the issues you have described. However, we will investigate this matter further and take appropriate action to address any issues that may have arisen. Thank you for your patience and for helping us improve our service.

Best regards, " :)

View solution in original post