Since its launch a few months ago, ChatGPT has garnered instant success that has opened up a notable streak around artificial intelligence. Conversational AI has been hailed in particular for the relevance of its answers in a wide variety of fields. Yet some of its limitations have also been pointed out. We have listed below 7 disadvantages of ChatGPT.

1. ChatGPT cannot respond to recent events

One of the flaws often mentioned about the chatbot is the limitation of its database, which has not been updated since September 2021. ChatGPT will therefore not be able to provide answers on current topics. Conversely, its competitor Google Bard (which will soon be deployed to the general public) is directly connected to the Web. An advantage that allows the AI ​​to access real-time information from the internet. Some alternatives to ChatGPT, like YouChat, also have the ability to access more recent information.

2. ChatGPT does not indicate its sources

To produce its content, ChatGPT draws from a substantial database, notably constituted from information present on the Internet (articles, websites, public conversations, etc.). The sources used are therefore multiple and unknown to the user. This lack of transparency is a barrier to ensuring the credibility of its content, or for users wishing to cite their sources (in the context of academic or journalistic work, for example). It is in this logic that Microsoft wanted to add this functionality to its new Bing integrating ChatGPT technology.

3. ChatGPT produces less personalized content

One of the inherent limits of artificial intelligence is the difficulty in producing singular texts or in expressing sensitivity. ChatGPT is no exception to the rule and offers content in an often cold and impersonal style. In particular, he tends to write standardized texts, which are similar to each other. If, in theory, the chatbot is supposed to be able to reproduce the pen of a person (a writer or a political figure for example), it actually has a certain difficulty in soaking up a style in a fluid and not stereotyped.

4. ChatGPT lacks sensitivity

An artificial intelligence cannot feel emotion or show sensitivity. At best, she can attempt to reproduce them. This leads ChatGPT to misinterpret certain subtleties of language such as sarcasm, humor or puns. This lack of sensitivity can also pose a problem when it comes to being critical. Indeed, ChatGPT sometimes lacks discernment in the hierarchy of information, by putting on the same level two discourses that do not have the same legitimacy. This tendency is accentuated by the neutrality of the AI, which seeks to nuance its responses as much as possible.

5. ChatGPT information is not always reliable

The vast majority of the time, ChatGPT produces accurate content, because it matches different information from its database. However, some users may have noticed inconsistencies in the chatbot’s responses. In particular, AI has produced from scratch sometimes very convincing false information (historical facts or quotes from authors), a phenomenon called “hallucination”. If the tool is likely to improve in the future, and to limit its number of errors, it is still necessary to verify the information it provides us.

6. ChatGPT’s knowledge of sharp topics is limited

Many specialists have been confronted with ChatGPT in order to assess their knowledge on very specific topics. In many cases, it has been noted that, while the chatbot is able to dissertate on the surface on a given subject, it is often unable to go into detail or talk about a sharp concept. These limits have been highlighted in particular in the medical field or in mathematics. Gross miscalculations were even made on relatively simple requests.

Even the CEO of OpenAI recognizes the limitations of ChatGPT in a tweet : “ChatGPT is incredibly limited, but good enough in some areas to create a deceptively large impression. It is a mistake to rely on it for anything important right now. It is a glimpse of progress; we have a lot of work to do on robustness and veracity. »

7. ChatGPT is not always accessible

In some cases, ChatGPT shows a long response time. It can also happen that it is not able to respond and shows the message “Network error” or “ChatGPT is at capacity right now”. These problems mainly arise when the request requires a significant effort from the AI, such as a long text or a complex topic. The error message can also be caused by too much concurrent traffic on the servers. Note, however, that OpenAI has taken the measure of the problem by offering full and faster access in its paid ChatGPT plus version.