Today, ChatGPT is relied on by millions as a helpful resource, be it for work, fun, or education. But can you rely on this AI chatbot to always provide the facts? Does ChatGPT ever lie, or will it only give you factual information?

Where Does ChatGPT Get Its Information From?

In its training period, ChatGPT was fed data from sources across the web, such as government and agency websites, scientific journals, studies, news articles, podcasts, online forums, books, databases, films, documentaries, and social media.

Specifically, ChatGPT-3 was trained using an extensive database of information that amounted to 570GB of data. As stated in an article from Science Focus, 300 billion words of information were fed into the GPT-3 system during its development.

Something worth noting here is that ChatGPT has only been fed data that existed before 2021. This means that the chatbot cannot answer questions relating to recent events. ChatGPT also does not have access to the internet. The data it was fed in training is the only kind it uses to fulfill user prompts.

But is ChatGPT only providing you with the facts, or is some more ambiguous information mixed in with its responses? Additionally, can ChatGPT lie to you?

Does ChatGPT Lie?

While ChatGPT often provides truthful information to users, it does have the ability to “lie.” Of course, ChatGPT doesn’t decide to maliciously lie to users, as it simply can’t do so. After all, ChatGPT is a language-processing tool, not an imitation of a real, sentient person.

However, ChatGPT can technically still lie through a phenomenon known as AI hallucination.

AI hallucination involves an AI system providing information that seems reasonable or plausible but, in reality, is not true at all. In fact, AI hallucination can provide information that it was never fed during its training period. Alternatively, it can occur when an AI system provides information unrelated to the prompt or request. An AI system may even claim to be human in a hallucination event.

AI systems, such as chatbots, fall into the hallucination trap for several reasons: their lack of real-world understanding, software bugs, and limitations on the data provided.

As previously stated, ChatGPT can only provide information using data published up to 2021, which certainly limits what kinds of prompts it can fulfill.

One of ChatGPT’s big problems is that it can also fall victim to bias when giving users information. Even ChatGPT’s creators have stated that the AI system has been “politically biased, offensive,” and “otherwise objectionable” in the past. As reported by The Independent, ChatGPT’s developers are committed to tackling this issue, but that doesn’t mean it no longer poses a risk.

When asked, ChatGPT stated that the reasons it may provide inaccurate information include the following:

• Ambiguity in the question (vague, unclear prompts);
• Incomplete information provided;
• Biased or incorrect information provided; or,
• Technical limitations (mainly due to lack of access to current data).

So, ChatGPT itself stated there are scenarios in which it will not provide accurate information to users.

In another response in the same conversation, ChatGPT stated that “it’s always a good idea to verify any information [it provides] with other sources.”

Can You Trust ChatGPT?

Because it can provide false information, you clearly cannot trust ChatGPT 100% of the time.

You can lower the risk of an AI chatbot hallucinating by setting specific parameters about how it can answer you. However, there’s still no guarantee that some false information won’t slip through the cracks.

Because of this, it’s best to check any information that ChatGPT gives you, especially if you’re requesting information about recent events. Double-checking this data by referring to other sources can help you determine if ChatGPT’s right in what it is saying and can prevent you from making ill-advised decisions.

ChatGPT Is Useful but Not Always Truthful

Unfortunately, you cannot rely on ChatGPT to provide truthful, unbiased information 100% of the time. This AI-powered chatbot is undeniably helpful and can help you in various ways, but it’s always worth verifying whether the information it provides is factual.