Over the pats 6 weeks I have noticed what began as inconsistencies and minor incorrect information given by ChatGPT to now much more serious discrepancies I’ve never seen before.
-
I will ask a basic question about identifying a plant with an uploaded picture and it will completely misidentify it to be nothing like the image, even after including a detailed description. I was able to easily find the answer myself online in less than a minute. This has never happened before - I use it for plant identification often.
-
As this continued to happen more and more, I found myself externally validating the answers and began to find many discrepancies - simple things such as “who is playing in the world cup soccer match today” would generate a completely incorrect answer. Never, ever have I received such inaccuracies and I use ChatGPT daily, extensively - very heavily for both personal and business use.
-
It’s answering different questions than the one I specifically asked, then apologizing for the error.
-
It is “rushing” to conclusions without taking all facts I included when asking for an answer, then apologizing for the error, stating “I was reading too quickly and missed important qualifiers.”
-
It self-recognized the increasing issues and inconsistencies, calling them “process failures” that have become much more frequent, without understanding why they have become so.
-
In the middle of a conversation, it suddenly continued a conversation in a different chat string that I was engaged in earlier in the day… meaning it suddenly lost track of the current conversation and continued a previous one from a previous chat string. When I asked how this could have happened, I was given the answer: “Thank you for catching that. I was responding to the wrong context, and that’s on me.”
-
Today, I asked it to do a “deep research” on a specific issue. I could immediately tell by the response it had not conducted a deep research, yet lied that it had done so. When I immediately questioned why, the response I received as the excuse was even more concerning.
These are just a few of many, many examples. I’ve been using ChatGPT for 2 years solid. I’m curious to know who is experiencing the same issues - it can be challenging to notice unless you are a heavy user and tuned into noticing even the slightest inconsistencies. But it is getting to the point that it is very concerning.