It seems like after level 100, the grid size starts to expand outside the bounding box the code is looking at. The code completely missed the top and bottom rows. You'd probably have to do something to make it expand the dimensions at higher levels, but tbh I wouldnt bother either. At that point it doesnt really matter anymore. XD
Would probably need to search for the boxes themselves within the entire screen and not just the small box. Could probably do something to make some type of grid with the center coordinate of each box, check the color of each of them at the start of the level, record the ones that are white and then when they flip back click all the coordinates in the list for the ones that were white. Sounds easier than it probably is lol
@@Swindly_YT you can also use selenium to locate the boxes, and either use the built in click function or use the location of the element and feed that into pyautogui
@@moofurg Ill have to take your word for it lol. Im not very knowledgeable on python. I can code in it but I dont know all the packages and stuff it has. I am guessing though that that basically is a form of what I was thinking just already made right? Im trying to get an understanding of how it works.
Starting from centre and counting boxes up would work. After the first gap you know the spacing of two squares, so you know when to stop searching. You'd need to take into account whether you start from a square (odd number of squares) or from the gap (even number of squares), though, which may cause some problems
The reason it failed at the end is because the vertical spacing of the grid started increasing while the horizontal spacing didn't, eventually making it too rectangular to fit inside the square that the code was confined to at level 105. It does look like it was still correct if you omit the top and bottom rows that it couldn't see.
along with that it started to change the general area that it took up so it would've started not checking some of the squares on the outside for the size if it hadn't run into the problem you mentioned.@@NiceLasers
You should mention this channel on your main channel once in a while (aka every video because they happen every fewmonths or so). I only learned about this channel recently and it's so much great content. Also do a full 100% of all human benchmark tests on a new account when you are done with all the individual tests.
Once I get to level 10, it becomes easier for me to memeorize the shape of the blocks that are not turned white. And then click the ones that are not..not white. Using that method, I got to level 39.
seeing the python error made me marvel at the fact that we have any working programs XD also made me go from "maybe i'll try my hand at coding one of these days" to "Hell no i'm not careful enough to notice details like that"
Those bugs can be hard to catch. What some people do is try to refactor their code to limit the amount of nesting (number of conditionals/loops insode other conditionals/loops) with the use of functions. Don't think just because that bug looks trivial that you can't do it. We have all made and continue to make those mistakes.
To add to the other reply, other languages use different (and better) ways to nest code, mainly curly braces {}. In my opinion, it makes it easier to follow rather than tabs, because you have an actual block of code.
Python: Tabs vs spaces or your indentations are in the wrong level. C, C++, PHP, etc: You forgot to close the line of code with ; But yeah most of the time, the runtime or compiler will catch something, just have to go look at the line number it spits back at you and be like oh yeah, I see it.
I gotta wonder how much work you have to do to fix ChatGPT code vs. doing it from scratch yourself, lol… I’d love to see you do a full video comparing the processes
I did the human benchmark tests after watching his vids on them. I got 119 on the verbal memory last night and just wanted to brag about it. thanks guys
This brings back old memories of me trying to do the same lol. My (program's) record for visual memory is 51.2 points. The trick is to simply ignore the grid and save a screenshot of the whole playfield. When it's time to click, pick any white pixel in the screenshot to click on until you've done them all.
2:20 technically you can calculate the size of the entire grid when you have the width of the first square, you don't need to manually count it. you can do something like: size = floor(board_width / square_width)
You can try to use some browser automation frameworks maybe. Playwright, for example. If every square is an html element, then the task will be more or less trivial. And it works fast
just did that, no need for playwright even. all you need is a mutation observer to trigger class changes, save them to an array and then trigger touch event on them, don't ask me why touch event specifically, cus click didn't work.
I think this is the first Code Bullet video I’ve seen with a face cam. 👍 4:31 I’ve seen this in so many other videos. People swear in their videos but censor the subtitles. Kinda funny that’s how YT works. 😂
its still 18x18 and i think the zooming for recording was just the reason it looked wrong. u can see that hes slowly been zooming in for the past few levels but since he cut the footage from lvl 101 to 104, the next shot we got of 105 is more zoomed in with a jump instead although it does look like the bot missed the top and bottom rows too which is weird
I wonder how much harder it would be to make an ML version of this... Like the input is the screenshot, and the output is the coordinates of the white squares? So it would be like a multiclass-multilabel classification problem?
@@overthr0w138 idk, a fantasy adventure type increment game? 🤷♂️ As long as he makes it, maybe while streaming it, it would also be fun to see his taught process.
you could do this in autohotkey using just progressively incrementing wait periods that follow with a click on the position of the square where it was earlier. There isn't really much of a memory in it but if you were to delay the click the appropriate time, it would work. However, making it overlap will be a bit of a challenge, it is doable and much more complex things have been done with it to the point it's almost turing complete. Even if it was turing complete, its complete inability to handle decimal numbers correctly would strongly limit its success.
I had this as my minor project. But I only could get to coding the game. :(( I had to change to a snake game , which i dunno was difficult but I did that in pygame which was faster to cook up.
An audio note. Can you please not have the music become louder than the highest peak of the person speaking. As the volume increase is off putting and becomes too loud. As the human voice varies in volume but the music doesn't so it is way louder in comparison. Talking about 6:42 when its just music playing. Lots of RU-vidrs do this and its always annoying.
when the memory test matrix dodges out of the way of the ai bullet but then the ai decides to just shoot the legs where the test has no chance to move away
ironically, someone can just take a quick screenshot and cheat their way into much higher levels.... just hit the print screen key and shove the image into paint or something, cause it doesn't seem like there's a time limit