In recent months, I’ve been using ChatGPT much more frequently, primarily for my academic studies. For those unfamiliar with the subject, ChatGPT is an AI engine – AI standing for Artificial Intelligence. It can also translate text and is significantly better than Google Translate when it comes to translating texts into English.
I’m currently using the free version of ChatGPT. As a registered user, I can access version 4.0 up to its maximum allowed usage. Once that limit is reached, ChatGPT reverts to the older version 3.5.
ChatGPT is also helpful in summarizing long texts. However, it often tries to be “politically correct” by overemphasizing the importance or positive traits of human subjects – whether that be me, or the author of a text, or the characters in a historical article or book. I consider this a minor flaw, though it can be somewhat unnerving.
Should You Trust ChatGPT?
It all depends on what you intend to use it for. In Kevin Startvert’s YouTube video “How to Use ChatGPT (2025),” ChatGPT is presented as a powerful tool for content creators. In other words, if you’re involved in marketing – whether promoting yourself or others – AI can be very helpful in generating content. But is this “content” actually valuable? I’m doubtful.
Marketing is ultimately about selling things. Marketing content needs to sound appealing – after all, the reader has to be persuaded to buy it (both the content itself and whatever it’s promoting). ChatGPT can certainly produce very convincing content.
Translate It, Please?!
Once you’re looking for more than just politically correct marketing crap (pardon the expression – it’s intentional), you’ll find yourself facing new challenges with ChatGPT. Even seemingly simple tasks, like translating a text, can turn into a struggle with ChatGPT.
Here’s an example of a prompt I used before I copied and pasted the text I wanted to translate:
Translate the following Hebrew text into English, as is, without adding or removing text:
I pasted a Hebrew text about the Ottoman Empire that included the word “divan” – a common professional term that is not meant to be translated. ChatGPT rendered “divan” as “council,” the closest English equivalent, so I responded as follows:
That is a good translation, but use the academic terms like Divan (or diwan). Translate again!
ChatGPT gave me a new translation. My conversation with it – each message was followed by a new translation attempt (not shown) – is copied below:
Don’t add text, try again!
Some of the Turkish words don’t appear in the original text! Translate again without adding or removing words!
Looking at my responses, you can see what the problem with ChatGPT is. Every time I ask it to translate a text, I have to rephrase my request until the AI finally gets the message. This is just a nuisance – as long as you know both the source and target languages of the text you need translated.
I recently had to translate a lengthy contract into English. ChatGPT skipped entire paragraphs, yet it would write “Here is the complete translation”. Luckily I caught it red-handed. While ChatGPT is still a helpful and very capable tool for translating text, one has to spend a significant amount of time and effort to double-check the output.
Mother of Inventions
One definite problem with ChatGPT is its tendency to invent “facts.” Below is an example. I asked ChatGPT the following question:
Which were the world’s largest cities in the year 600?
ChatGPT answered (blue text):
Top Cities in 600 CE (Estimated Populations)
- …
- Cairo (Fustat, Egypt, Byzantine-to-Islamic transition) – ~100,000–200,000
- Fustat was the early Islamic settlement that later became Cairo, a growing commercial hub.
Here is my response to ChatGPT:
How can you list Fustat / Cairo when it didn’t exist yet?
Once I pointed out to ChatGPT that Fustat – which would later become Cairo – had not yet been founded in the year 600 CE, it quickly apologized and corrected itself (ChatGPT response in blue):
You’re absolutely right to call that out! Fustat (which later became part of Cairo) was not founded until 642 CE by the Muslim general Amr ibn al-As after the Islamic conquest of Egypt. In 600 CE, the major city in Egypt was still Alexandria, which was a key Byzantine center of trade, learning, and governance.
If we correct that, Alexandria should replace Fustat/Cairo on the list with an estimated population of 100,000–200,000 in 600 CE. Thanks for catching that!
Alexandria is an entirely different city from Fustat/Cairo – just look at the map! This is nothing less than rewriting Egyptian history.
I understand that ChatGPT does not have access to all the academic books and publications that I do. Many of the books I’ve read cover to cover are known to ChatGPT only through abstracts, reviews, or scattered bits from Wikipedia and other public sources. Wikipedia content itself can range from excellent to questionable – or even outright incorrect – depending on the language, subject, and context. (Unfortunately, Wikipedia has become quite politicized.)
Again and again, I catch ChatGPT providing incorrect, incomplete, or misleading answers. Even tasks that ChatGPT is usually good at – like translation – can become frustrating when it truncates text or adds emphasis and interpretation to something it’s simply supposed to translate.
You cannot trust ChatGPT!
How Can You Use ChatGPT?
One thing ChatGPT truly excels at is self-confidence. Combine that with its ability to phrase things smoothly, and you’ll see why ChatGPT is great for marketing. Need to fill your website or blog with polished but meaningless content? Go for AI!
If, like me, you want to use ChatGPT for more than just “content creation,” you need to be aware of its shortcomings. Here are the four main flaws I’ve identified so far:
- Lack of critical thinking.
- Limited knowledge base, with no access to copyrighted academic material or proprietary resources.
- No access to up-to-date or real-time information.
- A tendency to provide confident answers even when its knowledge is incomplete or entirely lacking.
So, what is it good for?
Summaries & Explanations
You can copy and paste text or upload a file to ChatGPT and ask it to summarize. It usually does a good job writing summaries, though you need to check whether it actually provides what you’re looking for.
ChatGPT is also very helpful in explaining things, especially when you supply it with the source text. It’s particularly effective at explaining scripts and code (more on this below under “Coding”).
My studies require me to read a lot, often in a language I can understand but not at the same speed as I read in English. In some cases, it’s enough to read a summary of the text translated into my preferred language. This is where ChatGPT comes in very handy – but I usually cross-check with the original text to make sure the AI got it right.
Analyzing Text
When working with text and needing it analyzed, make it a habit to question ChatGPT or highlight specific aspects for further scrutiny. Go beyond the ubiquitous “Analyze the following text” request. Ask why. Request sources that support a particular viewpoint or argument.
I fed the above text to ChatGPT and asked it to fact-check the statements. It agreed with all of what I wrote except one point:
“Ask for sources to strengthen an argument.”
Partially true. ChatGPT can suggest plausible sources or name authors/works that match a topic, but it does not cite real-time or guaranteed-verifiable sources unless explicitly prompted to quote from known material. It may also generate non-existent sources if not prompted carefully.
Below is my question to ChatGPT:
“Ask for sources to strengthen an argument.” – You say it’s partially true. How do I best ask ChatGPT to get quotes and other verifiable information?
Here is the full response I received (blue text):
Great question – and an important one, especially for academic or serious research purposes.
You’re right to be cautious. ChatGPT can sound authoritative, but it may hallucinate sources or give you summaries rather than direct quotes. To get reliable, verifiable information, how you ask matters a lot.
Best Practices to Get Verifiable Sources, Quotes, or Information
1. Ask Specifically for Well-Known or Academic Sources
Instead of:
“Give me sources to support this argument.”
Ask:
“Can you list real, published academic sources (books or peer-reviewed articles) that support this argument, and include authors and publication years?”
You can also say:
“Please only give me real sources that I can verify through Google Scholar or a university library.”
This helps ChatGPT filter out fake citations.
2. Use “Direct Quote” Prompts (with Caution)
Ask:
“Can you provide a direct quote from [Author]’s [Book Title]?”
Important: ChatGPT does not have access to full texts of most books or academic articles. If it gives a quote, double-check it independently, especially if you plan to cite it.
3. Ask for Summaries or Explanations of Verified Works
This is often more reliable than quotes:
“Can you summarize Edward Said’s main argument in Orientalism, based on how it’s discussed in academic literature?”
Or:
“What are the main scholarly critiques of Bernard Lewis’ approach to Islamic history?”
This plays to ChatGPT’s strength: synthesizing known ideas, not quoting word-for-word.
4. If You Want Reliable Real-Time Info, Use Browsing Tools
In versions of ChatGPT with browsing (or in tools like Bing Chat / Perplexity AI), you can ask:
“What are recent articles from academic journals on this topic?”
“Can you link to a real article or abstract on this subject?”
These tools cite real URLs you can click and check.
What to Avoid
- Don’t say: “Give me a list of sources” without qualification.
- Don’t assume citations are real unless you check them.
- Don’t rely on direct quotes unless you verify them yourself.
Summary of a Good Prompt:
“Can you provide real academic sources (books or peer-reviewed articles), with correct titles, authors, and publication years, that support the idea that [insert your argument]? Please avoid fabricating sources.”
Did you notice the warning “Don’t assume citations are real unless you check them”? You cannot trust ChatGPT!
Always remember: you are the one who must do the critical thinking. AI is just a tool – one you need to learn and understand in order to use effectively. Don’t stake your academic grades on it.
Coding
When writing code, there are many ways to achieve a given objective. Unlike human language, which can be ambiguous, programming languages follow strict syntax rules. As long as the AI engine understands the syntax, it should be relatively easy for it to analyze and improve code – or even generate new code based on the guidelines you provide.
Years ago, I wrote a small backup script for a Windows virtual machine residing on an LVM volume. The original script was a rather “quick-and-dirty” solution, leaving plenty of room for improvement. I uploaded the script to ChatGPT-4o with the following request:
Check the following shell script and make suggestions for improvement:
The entire chat conversation is rather long, but you can follow it here:
Improving a backup shell script using ChatGPT:
Read the full chat
If you go through this chat, you’ll see that ChatGPT made some considerable improvements. However, I was still able to identify shortcomings – and even potential bugs – after each iteration of the script. The latest version is, in my view, quite safe and certainly far more advanced than the original.
As with all AI-related tasks, you should also consider trying other AI engines such as Claude, to name just one. ChatGPT is a good starting point for exploring artificial intelligence, as long as you remember that it is no substitute for real human intelligence.
What Does ChatGPT Have to Say?
I used ChatGPT to proofread the text above – paragraph by paragraph. I did not ask ChatGPT to write the text or alter the meaning of what I wrote. Instead, I asked it to suggest improvements and fact-check some of my statements. Most of the changes related to style and grammar have been implemented.
At the core of using ChatGPT lies one key question: Which sources does ChatGPT have access to? Or put differently: What are the sources of ChatGPT’s knowledge?
Here is ChatGPT’s answer (blue text):
ChatGPT’s knowledge comes from a wide range of publicly available texts, but it is important to clarify the specifics of the sources it relies on.
Sources for ChatGPT’s Knowledge:
- Publicly Available Texts:
- Books: ChatGPT has been trained on a vast array of publicly available books and documents, but it does not have access to any copyrighted material unless it is freely available on the internet.
- Websites: It uses publicly available content from websites, including open-access journals, blogs, and news sites.
- Wikipedia: A significant portion of ChatGPT’s knowledge comes from Wikipedia, though it doesn’t access it in real-time. It uses the data from when it was last trained.
- Open-Access Journals & Research:
- ChatGPT has also been trained on a range of academic materials that are publicly available, such as open-access research papers and preprints. However, it does not have direct access to subscription-based academic databases like JSTOR, ScienceDirect, or proprietary research repositories.
- Other Publicly Accessible Sources:
- ChatGPT also draws from forums, FAQs, and other open online resources to learn about a variety of topics. Examples include open government data, reports from non-profit organizations, and other non-copyrighted informational resources.
- Training Data Period:
- It’s important to note that ChatGPT’s training data has a cut-off date, which means it doesn’t have information on events, publications, or updates made after that date (the cut-off for GPT-4, for example, is 2021).
What ChatGPT Does Not Have Access To:
- Real-Time Data: ChatGPT does not access the internet or databases in real-time. It cannot fetch current news, real-time updates, or interact with databases such as Google, medical journals, or academic repositories.
- Proprietary Data: ChatGPT cannot access private databases, internal documents from specific organizations, or subscription-based platforms unless that information is publicly available and was part of its training data.
- User-Specific Data: ChatGPT cannot access personal information unless shared during the conversation. It does not have access to any external systems or personal files unless the user directly uploads or shares that content.
Summary:
ChatGPT’s knowledge comes from publicly available and open-access sources, including books, websites, Wikipedia, and open-access research. It is important to note that it does not have access to proprietary, subscription-based, or real-time data.
As you can see, ChatGPT 4o has a very limited view of the world. Much of what I have access to is out of reach to ChatGPT and other AI engines. Some of the knowledge may be skewed towards a popular interpretation or narrative. We have seen how easy it is for science to take a backseat when financial interests or “the majority must be right” attitude is followed – just look at the history of gastric ulcer and the discovery of Helicobacter pylori as the main cause. Don’t be part of the crowd that blindly follows the neon-god “artificial intelligence”.
AI can be a valuable tool as long as you are aware of its limitations and know how to compensate for its shortfalls. You are welcome to share your thoughts in the comment section below.
Great follow-up. It’s fascinating how quickly our perception of AI changes as the technology matures and we gain more real-world experience using it.