I’ll tell you how to solve some challenges in artificial intelligence.
The first problem: Unfixable code errors:
You, as a user, don’t know anything about programming, so you copy and paste the code, and you might encounter an error. If the model can’t fix the error, it’s stuck in a loop of repeating the error, and sometimes a new error. All you have to do is take the code (the original one where the first error appeared), open a new conversation, enter the code, and then enter the error. The model will most likely give you the correct code. Note that when you get the new code and it actually works, you can do two things: take the new code and send it to Gemini with an explanation of the error so it doesn’t make mistakes next time, or give it instructions to fix the code so it
can fix it itself.
2: Interface problems (create an interface for a website or
application). Focus here!.
Now, one of the most famous Experiments in artificial intelligence are that you tell it “Make a tic-tac-toe game or a snake game” but from my point of view you are not talking to a human employee, you are talking to a model that imitates patterns, meaning it has no “taste” or “opinion”, even its opinion is just the most appropriate possible response, for example, if this is a ball or a square, it doesn’t know anything, but a ball has a percentage of 0.984645 and a square has a percentage of 0.916464, so it chooses the highest percentage. What is important is that the best model is the one that knows how to do the following: For example, the traditional tic-tac-toe game location is “Make a tic-tac-toe game”, but it is better to tell it “Gemini today we will make a tic-tac-toe game, but we will make it in stages or chunks” meaning that you normally make the game once, the entire code, but it is not always the best thing and modifying it is not always the best thing, so instead of making the house and modifying it, we make it from the beginning, for example, we will start with the background, then the background effects, then the game box, and so on. We can make it as a ToDo list" in this way This way, you can control everything, and fortunately, Gemini has 1M tokens, so this method isn’t a problem at all. This is what makes Gemini stand out. I expect that if they only focused on truly improving the model, not just “trendy,” it would be the strongest model. The important thing with this method is that you control everything. If you want ideas, you can go to websites and get inspiration from them. Even if you want to save tokens, you can tell them, “With every modification to the code, don’t write the entire code from scratch. Write the part along with the modification instructions.” You can take all of this to a new conversation, and the code will be compiled, and you’ll have saved tokens. However, you’ll face a problem (dimensions), which is that the code that will be output won’t always have the correct dimensions. You can solve this by using ratios as a basic solution, such as instead of ( XxY dimensions), the box can be 30% of the screen, for example. This is a basic solution, but I’m too lazy to come up with a more powerful solution.
- A false (ineffective) belief that the most important thing is the word "Consider yourself (a programmer, engineer, doctor, etc.)
Many people think that in order to come up with strong responses, you have to say, ‘Consider yourself a programmer for ten years,’ blah blah blah. This is all useless talk because it gives you the same response with a bit of hype, and your response might be better without it (for example, the X-O game, the step-by-step method is 10,000% better than any fake method).
- Forgetting context and the difference between implementation and discussion:
Any project or long conversation of more than 100,000 tokens, if there are some side points (which I adopted) or philosophical talk that plays an important role but comes in a flowing manner in the middle of the conversation, it will usually be forgotten unless you emphasize it. The solution here, and this is literally the best solution, is to (write down). You must always write down at least the titles. Name titles for your questions and put them at the beginning of the message as a title. Here, you can make a list of what you need with notes. In case there are small but important words that you don’t want to waste, you can consider it like a ticket in Discord. Any inquiry or A stage is a piece on its own, but the difference is that it is a large, organized conversation. Or if you do not want to write it all down, you use the evaluation method. Meaning, we have 10 different messages, but on the same topic, and the first five are the strongest. Your attention mechanism, the model, will remember them well. But for example, message number 8 has a side word that is very important to you as a user, but with your innate feeling, you know that the attention mechanism will not focus on it. So, you can put a “tag” in the message that gives it an evaluation (philosophical) (technical inquiry) (side question “strong degree, question weak, average, strong”) So, here you are not making the attention mechanism focus. You can make a whole list with artificial intelligence on these points, so it will not get lost. Another very big mistake that users fall into is raising the
request → building a plan (by the model) —> implementation .
The best strategy is: - Raise the idea —> divide it into modules ----> take each part theoretically and make theoretical decisions ----> finish all parts ----> implement the parts and Editing
This is the best way. For example: Don’t say: “Game, we’ll start with the background. Come on, I want it to be black, and surprise me with effects.”
Instead, say: "Gim The first stage of the “Modules,” assuming you created them, is the background. I’m looking for specific colors that are easy on the eyes and fun. Give me your suggestions. As for the effects, I think we can get inspired by some famous backgrounds, for example, those that take on the character of the game, such as “Tick Tock.” Since it’s a tic-tac-toe game, can we add moving Tic-tac-toe effects in the background? What do you think? I want your opinion, and you can edit and improve it. (For your information, Gemini can give you strong suggestions, but the spark of the idea must come from you first.)
So, these are the things I had in mind, and I wanted to share them with you.
##Sorry if the words weren’t very clear, organizing the article is tiring.