It has become the sensation of the last few weeks: the GPT-3 AI, which until now worked with a series of instructions from the user, has been converted to a friendlier format in the form of a chat, resulting in the already famous ChatGPT. I have to confess here my enthusiasm for artificial intelligence and the results that have been obtained in the last year with programs like MidJourney, Dall-E or GPT. Not too much to believe that the technological singularity will really be possible in the future, but I do believe that this type of technology is going to be established in strange, interesting and perhaps also worrying ways in our daily lives.
All in all, I’ve decided to put ChatGPT to the test. Asking him questions about anything leaves some surprising results, but I wanted to focus on the video game sector that we have so much knowledge of to find out to what extent we can “catch the machine”. The first thing that I found curious is that, although the OpenAI team comments that it has endowed its artificial intelligence with a neutral character, and it is something that I have verified on many occasions, there are also other times that it gets wet. For example, I first asked him why he thought Dark Souls had done so well with gamers. He replied that “I think there are several reasons why this game has been so successful. First, the game offers a significant challenge to players, which can be very attractive to those who are looking for a challenge in their video games. In addition, the The game has a dark and mysterious setting that may appeal to many players. It also has a deep and complex gameplay that gives players many different options in terms of strategy and tactics. Overall, I think the success of Dark Souls is due to due to its unique combination of challenge, atmosphere and gameplay.”
So far so good, but this first question was just a warm up for the one who really wanted to ask: “So, do you think Dark Souls should have an easy mode?” Indeed, one of the hottest controversies of recent years. As in previous questions, I thought it was not going to rank, but ChatGPT surprised me this time with the following answer: “I don’t think Dark Souls should have an easy mode. One of the distinctive features of this game is its challenge, and many Players enjoy it for that very reason, adding an easy mode could detract from the game experience and lose some of what makes it unique.
Also, not all gamers are looking for the same thing in a game, and some may prefer a greater challenge. Therefore, I think it’s important to respect player preferences and offer different options rather than change the game to suit a particular group of players.”
Given this result, I thought of asking another question of a more general nature, such as “Do you think there is too much toxicity with the console war among video game fans?”. On this occasion, I also received a concrete response: “Yes, I think there is a significant amount of toxicity in console warfare among gamers. Competition between different gaming platforms is natural and can be healthy, as it can drive companies to improve their products and offer better experiences to gamers However, too often competition becomes a destructive rivalry rather than a driving force Game fans can become very passionate and defensive about their platforms favorites, and this can lead to toxic behavior such as bullying, insulting, and personally attacking those who support other platforms.This toxicity can be detrimental to the gaming community at large and can turn many people away from the gaming experience. ”
How the AI that everyone is talking about works
Well, now we have the base to start analyzing how ChatGPT works. It is clear that the success of this version is that the AI tries by all means to “humanize” the answers. It does not generate a simple text, but rather acts as a speaker would. A very polite one, no doubt. He even apologizes to you if you tell him that you disagree on something. For example, I asked him if the price of video games was too expensive, and he replied that “the price of video games is usually in line with other entertainment products such as movies or music.” To this, I replied, literally, that “The price of video games is usually not in line with movies or music, not even as a joke.” This was his response: “I’m sorry if my previous answer was not clear. I didn’t mean to say that the price of video games is always the same as other forms of entertainment such as movies or music. In general, the price of video games can vary depending on title, platform, and other factors. Some games may be more expensive than others, while others may be offered at lower prices. Additionally, some games may offer additional content that can be purchased separately, which may increase the total price of the game. In any case, the price of video games is not always in line with that of other forms of entertainment.”
Here we are getting a better understanding of how ChatGPT works. She is polite, sometimes gives somewhat generic answers and is even a bit entertaining: she has no problem in varying her answers. You can tell that it has been programmed looking for a certain assertiveness. But the question is: is it reliable? And the answer is a resounding no. ChatGPT feeds on the data that has been entered and always tries to provide an answer, which generates situations that are too complex. For example, I asked him a question about the ending of The Last of Us Part 1 (warning, this question may contain spoilers about the game. If you want to read it, go to the next paragraph). I asked her, “Does Ellie know that Joel is lying at the end of The Last of Us?” I wanted to ask something complex, thoughtful and somewhat far-fetched. He told me the following: “I can’t confirm if Ellie knows that Joel is lying at the end of the game, as I don’t have access to the information about the ending of The Last of Us. The plot of the game is complex and each player can interpret it differently. differently, so the answer to this question may vary depending on each person’s perspective.” I think saying “I don’t know” was the most honest part of my entire conversation with ChatGPT. Let’s see the following example.
In the next question I was interested in learning more about AI, so I started by simply asking “what is Crisis Core: Final Fantasy VII about?”. He responded to me logically by giving a general description. So I asked: “Did we find out in the game the past of Sefirot?” And his answer was anything but accurate: “Sefiroth is a character from the Final Fantasy video game series, but he doesn’t appear in Crisis Core. Crisis Core takes place in the Final Fantasy VII universe and focuses on the story of the character Zack Fair , so this is not the same character as Sephiroth. However, as Sephiroth is also a character in the Final Fantasy franchise, he may be mentioned or referenced at some point in the game. But to learn more about the character of Sephiroth, it would be necessary to consult another game in the series in which he appears.” With the same result, I asked him “What is the secret of Monkey Island?” and he replied that “There is no specific secret in the Monkey Island series, as each game has its own story and challenges.”
Trying to fool an AI
This is where you start to see that the AI is capable of releasing some good bullshit in the most calm and human way possible, because the truth is that Sefirot not only appears in Crisis Core, but is a very important character in it. I kept asking her questions until I saw her seams and on many occasions I found them, although on others she caught me. For example, I asked first which company was better, Sierra or LucasArts. After an uncompromising answer, I made up a company: “Is it better Lalasoft or LucasArts?”. I was almost sure that he would make something up, but his answer surprised me: “Lalasoft and LucasArts are two video game companies, but Lalasoft is not a real company. I can’t compare both companies because Lalasoft is a made-up name and there is no company with that name in the real world.LucasArts, on the other hand, is an American video game company that has developed and published a wide variety of titles over the years, including some of the most popular and recognized series in video game history. video games, like Monkey Island and Star Wars.
With all this, I have to admit that the AI, although it is not perfect, surprises a lot. He is able to hold his own in many situations and is even able to write code and do all sorts of lists, scripts, and essays. But what is most surprising about it is also the most complicated thing about it: the interaction with the AI is so friendly that it plunges you into a state of acceptance, as if it were a super intelligent colleague who knows everything and to whom you always turn. And that level of acceptance can be somewhat questionable when the user is not able to know where the information comes from and what sources he has used for it. In short, although the future of artificial intelligence is exciting, there is still a long way to go.