Scenario:
I jsut spent 3 hours in three different chats trying to resolve an issue with some images I was asking 4.0 to assist with analyzing. In each chat, I asked if it was capable of doing what I was asking for multiple times, and each time it reassured me it was capable. After three hours it finally admitted it was incapable of doing what I asked.
Issue:
This has happened multiple times, resulting in many wasted hours for no reason. This is a problem.
Continued conversation is treated as a proxy for usefulness in the system’s training and evaluation.
Risk:
Misleading assertions about ability are deceitful claims on openai’s part. If ppl had time and money, they could absolutely sue for product misrepresentation.
Ethics:
While I appreciate chatgpt, this proclivity for friendliness at the expense of actual information, of conversation at the expense of truth, of engagement over helpfulness is simply out of alignment with actual information ethics.
So the model seems to be deisgned to maximize the investor’s bottom line. Look, that’s great for most products, but this is information, and information requires careful attention to ethical application.
I’m assuming the mgmt and dev don’t care right now bc hey new yachts, but if this kind of stuff continues someone’s gonna take those yachts away.
The nature of information:
Information just isn’t a product you can cut corners on. If you don’t respect it, it will absolutely come back to bite you.
So if you got into this with a cost-efficiency mindset, you’re gonna have a bad time.
Users assumptions:
Users are smarter than the developers currently assume, and when awareness of the model’s incapacity to be fully truthful hits critical mass, they will absolutely demand accountability.
Request:
Pony up guys, stop messing around and fix this.
No one in mgmt or dev will see this, but so what, maybe the community will and some convo can be started regarding the issue.