-
This thread is just the tip of the iceberg.The people ahead of the curve aren't Googling for answers — they're already in here, having the conversations you haven't found yet. DealerRefresh is free.Get the full picture →
chatGPT told me that a "Can you" wasn't a yes/no question. It unconsciously made a mistake.
- Thread starter Carsten
- Start date
- Apr 29, 2011
- 527
- 285
- Awards
- 7
- First Name
- Chris
Gotta be detailed with your prompts! Ask it, "how many r's are in strawberry". With the free version it was returning 2. But if you prompt, "how many occurrences of the letter 'r' are in the word 'stawberry'"? it will return the correct answer.
- Apr 7, 2009
- 5,081
- 2,488
- Awards
- 10
- First Name
- Joe
ALL LLMs hallucinate. Like @Chris Cachor says, you need superior prompt engineering, or 'agentic prompts' to audit itself.
My experiences are: "The more complex the task, the more it hallucinates".
GOOD NEWS: CHATGPT 5.0 (strawberry) is coming very soon (this month)
My experiences are: "The more complex the task, the more it hallucinates".
GOOD NEWS: CHATGPT 5.0 (strawberry) is coming very soon (this month)
I use the paid version.
I have a pretty decent settings for how I want it to answer because I hate how it is over helpful and spams me with pages on how to implement something.
This latest round of changes they made, it is like it's trying to be human. During the original exchange I asked it if I asked it a yes/no question and it said no.
Eventually, it might actually work. lol
I have a pretty decent settings for how I want it to answer because I hate how it is over helpful and spams me with pages on how to implement something.
This latest round of changes they made, it is like it's trying to be human. During the original exchange I asked it if I asked it a yes/no question and it said no.
Eventually, it might actually work. lol
✨ AI Highlights
- A user reports that ChatGPT incorrectly classified a "Can you" question as not being a yes/no question, prompting discussion about AI limitations.
- The thread reveals that Large Language Models (LLMs) frequently "hallucinate" or make errors, with solutions including better prompt engineering and waiting for improved versions like ChatGPT 5.0.
- The key insight is that AI accuracy depends heavily on how questions are worded and that current models are still imperfect but improving.
A user reports that ChatGPT incorrectly classified a "Can you" question as not being a yes/no question, prompting discussion about AI limitations. The thread reveals that Large Language Models (LLMs) frequently "hallucinate" or make errors, with solutions including better prompt engineering and waiting for improved versions like ChatGPT 5.0. The key insight is that AI accuracy depends heavily on how questions are worded and that current models are still imperfect but improving.
Replies
Views
3
1,817
Started
Last Reply
Latest posts
-
-
Automotive Photographer Rates 2026?
- Latest: CarlAutofusion
