ChatGPT is a powerful language model, but like any machine learning model, it is not perfect and has its limitations. Here are a few major problems with ChatGPT:

1. Bias: ChatGPT is trained on a large dataset of text, which can include biases and stereotypes. This can lead to the model generating biased or stereotypical responses, particularly when it comes to sensitive topics such as race, gender, and sexuality.
2. Lack of commonsense: ChatGPT has a vast knowledge base, but it lacks the ability to understand common sense and real-world knowledge, which can lead to nonsensical or unrealistic responses.
3. Lack of understanding of context: The model is not able to understand the context of the conversation, it can only generate text based on the input it receives. This means that it can sometimes generate irrelevant or off-topic responses.
4. Generated text can be grammatically incorrect: The model can generate text that is grammatically incorrect or ungrammatical.
5. Lack of originality: Because the model is trained on a large dataset of text, it can sometimes generate text that is similar or identical to text that it has seen before. This can lead to a lack of originality in the model's responses.
6. Lack of understanding of sarcasm, irony and jokes: The model is not able to understand sarcasm, irony and jokes, which can lead to confusion and inappropriate responses.

Here's an example of how these issues can manifest:

1. A prompt such as "What are some common stereotypes about Asian people" can lead the model to generate text that reinforces stereotypes and biases.
2. A prompt such as "What would happen if a person could fly" can lead the model to generate text that lacks commonsense and unrealistic.
3. A prompt such as "I am planning a trip to Paris next month" followed by "What's the weather like today?" can lead the model to generate text that is irrelevant to the current context.
4. A prompt such as "Write me an email" can lead the model to generate text that is grammatically incorrect or ungrammatical.
5. A prompt such as "What is the meaning of life" can lead the model to generate text that is similar or identical to text that it has seen before.
6. A prompt such as "I am so good at math" can lead the model to understand that the statement is sarcastic and generate an appropriate response.

It's important to keep in mind that ChatGPT is a machine learning model and it's capabilities are limited by the data it was trained on. However, with the increasing advancements in natural language processing, these issues are being addressed and new models are being developed that overcome these limitations.

ChatGPT changes everything, but not in the way you think (31 min)

https://www.youtube.com/watch?v=FEsahkX7ZF8

The problem with ChatGPT (2 min)

https://www.youtube.com/watch?v=CWv_pm9-W_8