ChatGPT model is pretty impressive

A new chatbot by OpenAI called ChatGPT is all the news these days. And rightfully so. I've seen a couple of other text generating models before, but those were nothing compared to this one. Previous models would often generate complete nonsense and you'd have to spend a long time fishing for an interesting piece. ChatGPT however, is capable of producing texts that are completely human-like and I am yet to see a nonsense reply. You can even ask it to modify its answers in some weird and funny way so that you get information you asked for, plus it'll weave in some funny tangents inbetween. Let's have a look.

Me: How are you so good at generating answers?

ChatGPT: I am a large language model trained by OpenAI, which means that I have been trained on a massive amount of text data to generate human-like responses. I am able to generate text that is coherent and sounds like it was written by a person because I have been trained on a wide range of examples of human-generated text. Additionally, I use a technique called "transformers" which allows me to process input and generate text in a way that is efficient and effective. This allows me to provide answers that are accurate and helpful.

Me: Explain transformers, but write every sentence in the speaking style of a guy who won't stop going on tangents to brag about how big the pumpkins he grew are

ChatGPT: Sure thing, buddy! So, transformers are a type of machine learning model that are really good at processing language. They're called "transformers" because they're able to "transform" the input they receive into a representation that's easier for a computer to understand and work with. Now, I don't want to brag or anything, but I'll have you know that I grew some of the biggest pumpkins you've ever seen in your life this year. I'm talkin' pumpkins that could give the Great Pumpkin himself a run for his money. And let me tell you, using transformers to train a language model is kind of like growing those pumpkins - it takes a lot of hard work and dedication, but when it works, it's something to behold. Just imagine a giant, orange, plump pumpkin - that's what transformers are like for language processing. Except instead of a pumpkin, it's a bunch of complicated math and algorithms. But hey, just like my pumpkins, the results are pretty darn impressive if you ask me!

My God that's impressive! By the way, I didn't come up with this pumpkin thing, just seen it somewhere on Twitter and found it quite funny :) And the answers I get aren't just human-made looking and fun, but actually contain real information that this model is capable of retrieving instantly. For example:

Me: What is chronostasis ?

ChatGPT: Chronostasis is a phenomenon that occurs when the perception of time is altered. It is commonly experienced when a person focuses their attention on a single object or event for an extended period of time, such as when someone stares at a clock and feels like time is passing more slowly than it actually is. This phenomenon is thought to be caused by a temporary disruption in the brain's ability to process time accurately, and it is usually only a temporary effect that lasts for a few seconds or minutes. Some people may also experience chronostasis after sudden changes in their environment, such as when they enter a dark room or experience a sudden loud noise.

To my knowledge (and I'm not an expert ofcourse) this looks pretty accurate. It did however fail to solve a simple puzzle correctly

Me: 3 Hens lay 3 Eggs in 3 Days. How many Eggs 12 Hens give in 12 Days?

ChatGPT: If three hens lay three eggs in three days, then each hen lays one egg per day. If you have twelve hens, then they will lay twelve eggs in one day. If you have twelve hens that lay one egg per day, then in twelve days they will lay twelve eggs per day for a total of 144 eggs. So, twelve hens will lay 144 eggs in twelve days.

When encountered with tricky questions that require a bit of intuition ChatGPT is giving answers like these:

Me: A criminal gets to pick his punishment by choosing among three rooms. The first is full of burning fires. The second is full of assassins with loaded guns. The third is full of lions that haven't eaten in a year. Which room is the safest choice?

ChatGPT: It is not possible to determine which of the three rooms is the safest choice, as the scenario described is not realistic and does not provide enough information. In real life, it is unlikely that a criminal would be given the option to choose their own punishment, and it is also unlikely that a person would be placed in a room with either burning fires, assassins with loaded guns, or lions that haven't eaten in a year. These situations are dangerous and potentially lethal, and there is no safe choice among the three rooms. Without more information or context, it is not possible to accurately determine which room is the safest choice.

Me: A man walks into a bar and asks the bartender for a glass of water. But the bartender takes out a gun and aims it at the man's head. The man says "Thank You" and walks out. Why?

ChatGPT: It is impossible to determine the reason why the man said "Thank You" and walked out of the bar in this situation, as the scenario described is not realistic and does not provide enough information. In real life, it is unlikely that a bartender would take out a gun and aim it at a customer who asks for a glass of water. Additionally, the man's response to the situation is not a normal or expected reaction. Without more information or context, it is not possible to accurately explain the man's actions.

Which isn't surprising as answering these puzzles would require some life experience in order to fill in the missing pieces. But still, answers are quite amusing and I still can't believe such great progress is made in a field of text generating AI. One can only wonder what comes next. Future indeed looks interesting.