The Dark Side of OpenAI ChatGPT: Thinking Traps and Potential Crises

OpenAI, the artificial intelligence organization invested in by Bill Gates, publicly released an AI conversational program called “ChatGPT” at the end of November, sparking a lot of discussion online and a sense of crisis among humans about being replaced.

ChatGPT took only 5 days, without advertising promotion or spending money on marketing, to push its registered user count to the million level, becoming an important milestone in tech history.

After all, the author does not have much research on artificial intelligence. Technical features such as GPT, BERT, PPO, SOTA, training, and models are all unfamiliar to me. In addition, OpenAI has not yet published its research literature, so let us look at ChatGPT’s capabilities from the user’s perspective.

Let me state the conclusion first: for existing experts and masters, ChatGPT is like a Kinder Surprise egg — novel, fun, and tasty, satisfying three wishes at once, and can serve as a work-support tool.

But for beginners just starting out, it is not necessarily good news. It is a dangerous road full of thorns. It is hard to say whether the massive knowledge ChatGPT possesses can make you succeed, but it will certainly require people to pay double attention. Because with one careless move, ChatGPT is the biggest rumor machine.

Can ChatGPT Replace the Google Search Engine?

Let us first ask ChatGPT a simple historical question: “Who won in a fight between Zhang Fei and Yue Fei?”

Let us see how ChatGPT answers: “Zhang Fei and Yue Fei were famous generals of the Three Kingdoms period, and both were great generals under Liu Bei. According to records, Zhang Fei once fought Yue Fei on the battlefield, but the details of this matter are not very clear. Zhang Fei and Yue Fei eventually became important strategic pillars for Liu Bei, helping Liu Bei establish his kingdom.”

??? So it turns out Zhang Fei and Yue Fei actually had a hidden, unspeakable fight in history.

The Pros and Cons of ChatGPT

The advantage of artificial intelligence is that it can process large amounts of data, find keywords from it, and reorganize information according to human requests. It reduces the time of complicated procedures, reduces human error, and improves work efficiency.

For example, writing press releases is its strong suit. Because the genre of press releases can actually be produced through a formula (or template), you only need to provide bullet-point key points and plug them into the correct formula, and you have a decent-looking draft — in fact, most clerical professions do this. Whether news reporters, lawyers, judges, and so on, they all have their own template libraries.

But artificial intelligence also has one biggest problem: its answers cannot become a model, at least not yet.

Even OpenAI itself knows this, so you will hear it constantly saying: “I am just a large conversational program. Don’t blame me if there are problems; if it messes up, bear it yourself…”

I like to use an example to explain to friends the remaining difference between humans and A.I.: “The human stream of consciousness is 1234567890, while artificial intelligence is 1Z34SG7B9O.”

When we seek knowledge, there is a very important premise: we need to obtain a “complete and practical” answer. The mental activity of human thinking is like moving from the big picture into details, then depicting the relationships between each detail one by one, and finally returning behind the fourth wall to re-examine whether the entire work is perfect.

If we use the Google search engine to find answers, each article is actually a highly complete work produced by human mental activity — content farm garbage articles do not count.

But if we use ChatGPT to find answers, a problem arises: they do not care whether their answers are correct and perfect. They are just in the process of collecting data and training models. Simply put, they gather a large amount of text, then copy and reorganize it, and produce an article resembling a template structure.

If you repeatedly ask it a question, you may even get contradictory results. And there are many obvious small errors that keep appearing in the lines of its answers, such as 56 not being divisible by 7.

When used in specific commercial settings, because the weight ratio of input data can be controlled, artificial intelligence can do very well. But for the myriad strange questions on the internet, it becomes like speculation, just finding a similar answer for us.

The danger lies here, because during the Q&A process, our attention is already drawn directly into the details. For experts, it is relatively easier to notice what is missing and then supplement or organize the priority relationships themselves.

But anyone who has helped revise articles knows that rather than revising an article written with a wrong context, it is faster to delete the whole paragraph and rewrite it.

For beginners and students, because they simply cannot notice what key points are missing, and they think they have found the answer, they may be too lazy to think further and not look for materials to repeatedly confirm. At this point it is very much like a programming bug: unless you actually execute it (an exam, a proposal), it is very hard to discover that you are wrong. It might even be that the time spent correcting errors wastes even more energy.

But it is undeniable that at this stage artificial intelligence is only a transitional period; it is continuously learning and evolving. This is also why OpenAI made it freely available to everyone — any Q&A we do on it will continuously become nutrients for its growth.

Finally, I asked ChatGPT another question: “What is the difference between sesame oil and Sichuan peppercorn oil?”

It answered: “Sesame oil is sesame oil mixed with cilantro oil, so it is fragrant. Sichuan peppercorn oil is sesame oil mixed with Sichuan peppercorn oil, so it is numbing.”

🙄️🙄️🙄️🙄️🙄️🙄️

Next time on Facebook, if you see someone claiming they have not opened Google for several days, sincerely consider that person’s skill at bluffing — it may be as artificial as AI.