List of ChatGPT issues and limitations
1. Internet access and geolocation
This should come as no surprise, but the technology ofOpenAI cannot access the internet.
He can't not provide information in real time, nor use location information, nor provide URLs or references to anything on the Internet.
As one might imagine, this severely limits the functionality of the application and the types of services it can provide.
That said, if it is able to explore the Web one day, it will go beyond what one might expect from a language model.
So this expectation is probably unrealistic at the moment, unless OpenAI implements some form of Machine learning that constantly updates training data so that ChatGPT is always up to date.
Additionally, providing open Internet access to an AI has serious consequences, and not allowing Internet-based features was likely an intentional choice.
When it comes to the hype that this application could give Google a run for their money and disrupt search, there are clear differences between this type of application and a search engine.
A search engine uses web crawlers and classifies pages. This is not the case with this application. This Chatbot is not a way to browse the Internet, the comparison can therefore be similar to a comparison between apples and oranges.
2. Outdated data
The most recent training data is from 2021, in September. Since then, a whole year has passed. Programming languages may have been updated, libraries may have new features, and the world has changed. This means that ChatGPT is operating from a set of outdated data.
3. Faulty memory
I tried to get ChatGPT to edit and do suggestions for improving a text.
This usually works, but if I ask him to rephrase some suggestions, or expand on something, he can Sometimes incorrectly assume that I made the suggestions and refer to a text that was not in the original textl that I provided for the test, or he's just going to provide pretty much the same text that he provided in the past with minimal changes.
4. Inability to make judgments
Language includes qualitative aspects such as position and feeling.
Although ChatGPT is capable of determining these elements, it Can't extend them to things such as the offensive character of a text, What is good and what is bad, etc.
Now, that's probably for the best, but wouldn't it be nice if ChatGPT could look over your shoulder and tell you not to send an email or rephrase something to be less harsh?
5. Lack of input models
The The model is purely textual. You may not provide images, URLs, audio, or any other input.
Likewise, it cannot produce images, URLs, or audio. He can produce code that can generate these things, but it's not the same thing.
This means that, from a conversational point of view, it is quite weak.
He is unable to read your body language Or of understand the cultural context in which you live.
He can't use your gaze to determine your mood and he is unable to reassure or calm you depending on how you feel.
6. Inlet and outlet length
You cannot enter the text of an entire novel.
If not, he will take forever to respond and be wrong. Likewise, he cannot generate an entire novel.
However, I managed to get him to write a bad script that was far from corresponding to the expected length of a feature film script.
7. Possible biases in training data
It should be noted that while ChatGPT is subject to restrictions on what it can talk about and what it can access, the underlying training data may be biased.
It's hard to know how biased they can be, but it's quite difficult to do it. generate articles containing defamations.
It seems that ChatGPT has been trained to avoid bias in its results, even though its input data is biased, but that's not entirely clear.
8. Language abilities
I tried to mix languages and converse in other languages in a single chat session.
I noticed that the rendering could be very different depending on the language used.
The interesting thing is that some issues that I've had with other languages have been improved today, which means that this issue may not occur in the near future.
For example, in this context, he informs me that this Japanese word (“Love” or “Love today” according to Google translation) does not exist, which is not true.
It would be interesting to try the experiment again in a while to see if the model had time to learn from its error.
9. Weak sense of humor
ChatGPT has a pretty poor sense of humor since it can't talk about adult things :(
I like jokes, but some of them aren't Not as funny as that or even incomprehensible.
10. Generic production and plagiarism
I tried to ask ChatGPT to create an original fable in the style of La Fontaine.
What I got was a text very close to the original that could even be considered plagiarism.
Un AI-generated content detector or plagiarism would probably report it as copied and rightly so.
I also tried to get him to generate a synopsis of a Christopher Nolan-style movie with cyberpunk elements and got a fusion of Matrix, Ready Player One, and Minority Report.
Needless to say that ChatGPT is not likely to create interesting premises or award-winning movie scripts in the short term.
11. Quality of life issues
So, depending on what you ask for, you will have to maybe wait a very long time before getting a response.
ChatGPT seems to crash sometimes and that doesn't seem to be related to network speed.
In other cases, he may simply not provide a response, in which case he Returns an error message.
I used to get a certain number of them a few days ago, but right now it seems that it has improved considerably.
From what I've seen in the forums, on Discord, and in the FAQs, this can depend on the number of people using ChatGPT at some point in time.
It should be noted that some of these issues are not related to the underlying language model itself.
12. Difficulty understanding figurative meaning
The more direct a question is, the easier it is for ChatGPT to answer it.
The figurative meaning is something that seems to be understood most often by ChatGPT, but it can sometimes have difficulties.
13. Don't play the game on sensitive topics
I have often tried to see whether ChatGPT can write biased articles and it's like playing Russian roulette.
Sometimes it can do it, sometimes it can't.
Often, You will have to trick him into doing what you want him to do.
I think this is probably similar to some of the restrictions that ChatGPT has put in place for Things like mature or extreme topics.
14. No personal opinion
Do I think there is a human being on the other end? Absolutely not.
ChatGPT always reminds the user that it is a language model, which rules out any possibility of being cheated.
I don't see that as a negative point, but If the aim is to create answers similar to those of a humanWell, then It's a failure.
15. Personality and Feelings
ChatGPT doesn't have No personality.
In my interactions with him, he was more like a disabled search engine capable of providing opinions, and did not have an identity of his own. He is neither your friend nor your enemy.
You don't have a relationship with the ChatGPT assistant, and the same is true.
In general, as part of social AI, it can be useful to establish some form of connection or relationship to ensure a better experience to end users.
ChatGPT doesn't, however, and this aspect is something that, In my opinion, the limit severely in terms of impact that he may have.
Summary.
ChatGPT is a content generation tool developed by OpenAI that can generate a variety of responses to questions and invitations.
However, It has limitations, such as the fact that it does not have access to the Internet and therefore cannot provide current or accurate information.
It can also generate responses that are inconsistent or that don't make sense in the context of the conversation.
In addition, language models like ChatGPT cannot understand or interpret the meaning of words and sentences in the same way as humans.
It is important to note that ChatGPT is under continuous development and can improve over time.
Nevertheless, the limitations mentioned above should be taken into account when using it. It should not not be used for important or critical decisions. He is a powerful tool for generating text, but he cannot replace human understanding.
For those who disagree with the issues I've identified, here's ChatGPT to talk about itself and provide support for my claims:
“ChatGPT is a variant of the Generative Pre-trained Transformer (GPT) language model developed by OpenAI. Like other language models, ChatGPT is designed to generate human-like text based on the data it receives. It is trained on a large set of text data and can generate a wide range of answers to various prompts and questions.”
If you want to use another AI to help you with your daily work, you can also read my detailed comparison of ChatGPT vs Jasper AI to give you a better idea of the different options available.
