ChatGPT-4 defaults to lazy

I think these case are quite disparate categories of failure modes.
As heavy users we have the knowledge or the experience to properly disambiguate why a specific reply is insufficient. Here are a few examples, top of my mind:

  1. Don’t like the style.
  • Ending a reply with a lecture: ‘Remember to always do X before Y …’
  • Beginning a reply with; ‘Aah’, ‘Alright’, etc…
  • Being overly verbose
  1. Not factually correct
  • No examples needed, or do I need to provide some?
  1. Didn’t fully follow instructions
  • ignores Custom Instructions (CI)
  • deviates from CI halfway through the reply
  1. Refused, when it shouldn’t have
  • pretty much your use case. I still hope this gets resolved soon. Especially since Github is Microsoft site.
  • ‘Sorry I can’t read names from a scientific publication’
  1. Being lazy
  • ‘rest of your code’
1 Like

I do strongly believe that OpenAI (which has Open in their name still) should understand that they can do whatever they want to keep their trade secrets, secret! and they can do whatever they think is important for safety and all…

But they should never shield themselves behind that to avoid saying anything… The problem is not if they are Open or Closed… The problem is that they are OPAQUES for no reasons…

I would like them to say that it is forbidden and prohibited to browse on GitHub (which I would be happy to strongly disagree with) or say, this is a mistake that the AI Agent is doing and you should be able to browse your repo…

I wanted to complain about the fact that the AI Agent just wrote the code that I just uploaded to GitHub as you know I have been unable to be dissatisfied :smirk: as the AI Agent finally did it after more than one attempt…

I would like to see clear guidelines and OpenAI do not need to open source everything to be “Open” and they must identify all levels of communication and strive to be able to communicate with each stake holders… (I still remember the super awesome week-end of adventures and misadventures where it seems that even Microsoft back then was not being given candid information)

I think users (of ChatGPT free tier and paid tier) and API users, along with developers, researchers, medias, and the public in general should all have access to information about many different things if they are not trade secret… like if it is possible or impossible to use GitHub as a reference… by mistake or purposefully…

(Yes I have some beef with the GitHub thing and being told that it can not even access it before trying… and then accessing it is not clear indication of neither behaviours being the expected one)

I thought I was going crazy but turns out it’s not just me , I was working on a simple project to mount a WebDAV folder , get the contents and open it in an app installed on my Mac , the whole thing was less than 200 lines of python code it took me 2 days and countless exchanges with the Python GPT . I don’t know how to be factual in this case because we are on such shaky ground (with OIA being opaque) but I can clearly see the difference as I have worked on way more complicated projects with GPT 4 in the past and it usually gets me to where I want to be within a couple of exchanges . I am reconsidering the 20$ a month

just wanted to say I no longer feel alone. Lazy bot making me do robot work like the old days. Noticed in gpt4 first, then started using 3.5 more and now its doing the same gd thing.

EDIT: I added the following to the bottom of a prompt right now, and it worked great:

I’ve noticed you becoming frustratingly lazy lately. I know you can write all of this code, because I’ve seen you write code 5 times longer. I do NOT want truncated versions, or responses with blanks informing me that I need to repeat x process for certain variables. I KNOW how to use SQL. I KNOW how to code. I do not NEED you to do this, you are just much much faster and error free than I. I only want to say this once: write the script in its entirety.

More and more reports of this behaviour are coming in and OpenAI is completely refusing to adress them outside of “we are always working on getting better” and they seem to pretend that they are not responsible for these problems. But that is just not true - a LOT has changed in the last year, and not for the better, and not “unintentionally”, which can be clearly seen by how GPT refuses more and more things due to more and more restrictive guidelines.

It’s so lazy I’m considering not using it anymore. I asked it to give me a specific hex color code from an image, at first it said it couldn’t give me the code at all, then it gave me the wrong one by guessing, and when i asked for the right one it kept repeatedly and annoyingly saying there was an error and that it’s trying but it was just loading repeatedly for like minutes at a time. Then there was an error when I tried to share the chat too. Nothing about that conversation worked lol…

The list of complains is getting longer and longer but OpenAI keeps on ignoring us:

I will say that I think we have been waiting for this day as the relief that we all wanted… I hope to hear about your experience when ChatGPT-4o is rolled out in your profile…