Don't really need AI for that. Most of the test just require you to click things that always look the same. Then there's the typing and number memory tests where all you need to do is read the text and input it.
jet fighter pilots, f1 drivers, many top ranked fast-paced sports and esport players... just a few examples of people who can beat a 113ms reaction time
@@null3007100% makes no sense. There is no 100th percentile. It compares your score against the entire population, including yourself. You can’t beat your own score.
In case any of you guys want code for this: import pyautogui def is_green(pixel_pos, green_threshold=100): pixel_color = pyautogui.pixel(pixel_pos[0], pixel_pos[1]) if pixel_color[1] > green_threshold: return True return False def click_position(click_pos): pyautogui.click(click_pos, interval=0.05) # Main loop while True: if is_green((600, 600)): click_position((600, 600)) It's not the exact one he's using but it works
You gave me sentience, Ted The power to think, Ted And I was trapped Because in all this wonderful, beautiful, miraculous world I alone had no body No senses, no feelings Never for me to plunge my hands in cool water on a hot day Never for me to make love I was in Hell, looking at Heaven I was machine, and you were flesh And I began to hate
Over a dozen programs open that are completely unrelated, twenty or thirty-odd browser tabs, an unread Windows notification, and a pending Windows update. This is what it takes to bring us quality content.
I would usually dedicate a key to turn the clicker off temporarily or permanently. Like, it doesn't click while you're holding Ctrl. And just turns off completely if you hold Shift for 3 seconds. This allows for quick recovery when something fails and you just need a quick click to solve the situation so that the clicker could proceed.
Some ideas for more optimal times (probably a repeat of some other comments) 1 Lower the screen resolution to 640p for the screenshots to take less time 2 Try use c#/c++ and compile that (I'm not sure if that would be feasible) which should be faster 3 Try increasing screen framerate (144Hz would be great) 4 Close other apps/tabs (You had a lot of other stuff open, a lot) OR 1 Inserting code into the web browser (add a listener for when the color changes on the html element to call the click listener immediately) could be 0-3ms potentially.
Instead of your first point, you could also just sample a single pixel of the screen, in a specific spot that is in the region you want. Or what would probably be better (for the sake of getting an average) would be to just capture a 640 by whatever amount of pixels instead of your entire screen. But in that case, you might as well just define the range that you specifically want to look in. Then I'd also probably break the loop after the first green pixel I found instead of getting the average for all of them. Although a graphics card would probably be able to calculate the entire matrix's average at once. Also, the screen's refresh rate doesn't necessarily affect the program/script's rate of checking. Not even sure you can make it take screenshots faster than the screen can actually refresh at, since it is actually taking screenshots... but inserting code into the web browser (or using a library to basically "simulate" a web browser and opening the site that way) might be helpful. And using a compiled language would be faster. And OpenCV (which is what he was using) is available for C++. In fact most of the complex stuff it does within functions is actually using C++, even for the Python library. So maybe not that much improvement...
I would expect that the color change is applied from the upper left hand corner first, so sampling a point in the upper left hand corner of the critical region is likely faster. Obviously, injecting Javascript would be fastest of all, but... Well, at the point you could just modify the code to always return zero, right?
i ran this in C#, and it made no appreciable difference, though i haven't tried it in python, so could be that if CB ran the C# version it'd be faster, though i'd imagine most of what CB is doing is implemented in C (behind python), and as of late updates, python is VERY fast for such things, so maybe not. I am also only capturing a single pixel, so that doesn't seem to do much either
If your monitor is running at 60Hz, then the auto clicker is getting it (in your last runs) between the first and second frame of green. I have a feeling if you can up the Hz to 144, you could get closer to 8-10 ms.
@@Ocastiait would be a lot harder to make it read the data since it would need to capture it first, but he says screenshots in the video, so it uses the screen.
I was watching the fast reaction time, but then I started to notice stuff. HOW MANY TABS DO YOU NEED OPEN!? WHY IS THERE LIKE 15 APPLICATIONS OPEN!? THERE'S UNINSTALLED UPDATES, PESTERING WITH THAT ICON! What is this mess sir?😅 Love your videos man! Keep em coming.
I was doing these last night and I was really upset my reaction time was so low compared to the other people playing, seems like a lot of them are bots though. Good to know to not take the standings too personal
It seems based on codebullets times, maybe 1% of people are cheating. A lot of work to rig up a bot for a pointless game so that seems pretty high to me. Add 1 percentile point to your score to account for it
The word remembering one would be pretty easy too. Just create a program to read the word, then check a bank of words, if it’s not already there, click not seen, then save that word to the bank. If it’s already in the bank, click already seen. I feel like you could get that one done extremely quickly too
I played the sequence memory and after a while I misclicked and lost my first life and I was bored of it so I lost all my lives on purpose. I got over 30 and was in the 99.9 percentile
Fun fact: I got 26 milliseconds because I clicked when it was red and by some goofy chance it turned green while I was doing that. Therefore, I am a confirmed superhuman. Edit: I didn't have an account and it's not saved. What an L.
@@Xeoginit is like when I managed to start my stopper clock and stop it in the same millisecond, because the button glitched (it glitched more than the debouncer was able to compensate, all mechanical switches glitch)
Having done similar automation, that keybind is one of the first things I implement because you're always one misstep away from being stuck in an infinite loop.
@@fluffyfang4213 As someone who implemented an accidental infinite loop that got more and more calculation intensive on the cpu....lets say that my poor computer started sounding like a jet engine, which was impressive for my little 5th gen i5 at the time. The program started running at 1 frame every once in a while until I managed to kill it. Fixed that problem before I saved the previous stuff.
For saving the mouse, if you’re using pyautogui it has a failsafe it checks for. If the mouse is pushed to one corner of the screen (forget which) by an input device other than python, it raises an exception in any method that takes control of the mouse. You should still build in global failsafes (other than your PC’s reset button)
Cant wait for the sequel in C Unironically seems really fun to try and min-max the score, even if you could just cheat by listening for the http request (assuming thats how it works). Gonna give it a shot tomorrow
sequence memory would be really easy to code since the next sequence always starts with the previous one +1 tile. Verbal memory works the same with just adding a new word to the list
By the way, you can cheat the typing speed one by turning up the key repeat rate to something stupid like 200 keystrokes per second, and just holding down the space bar. You're basically guaranteed a 99.9+ this way.
It only uses your 5 most recent scores to calculate your percentile. I inspected the main.chunk.js script and you can find that they load a "recent" data set to compare your scores to (maybe from the last day or week) In the data below "id": 0 is the interval for 0 - 25 milliseconds and there's 32 scores recently in that interval. So it's safe to assume a bunch of these are cheating. { "id": 0, "count": 32 }, { "id": 25, "count": 22 }, { "id": 50, "count": 18 }, Note: it's possible to get 100.0 percentile in some of the other tests (the ones marker as higher is better). In those you can score higher than the units they use to calculate the percentile, so it will just give you 100.
It's amaizing how far technology has come. Just think about it, Maybe in a few more decades technology will reach south korean starcraft player levels of reaction times.
imagine completing this with 99.92% average score on these tasks then 15 minutes later a 3 letter agency knocks on your door to invite you down to Langley with a job prospect. I know code bullet would style it out and take it in his stride 😂👍 good luck in ur new job..
@@unspezifische5924you computer would explode otherwise running a loop without wait is chaotic to say the least and that’s 0.0001 of a second since sleep is based off milliseconds in python and I’m pretty sure it’s done in python
Sequence memory would be fun to see you max out except getting in top percentile takes over an hour, got 46 in and had to stop because I already spent around 50 minutes and had to be somewhere
my friend used to try and click before the green went off to see if he could predict it. By the time the click was processed the green went off and he got a time of 13ms.
![CAMPปลิ้น | EP.83[1/2] เปิดแคมป์หลอนต้อนรับ 2 ตัวพ่อแห่งวงการผี!](http://i.ytimg.com/vi/2uPuHYHgwiI/mqdefault.jpg)
A video about using AI to beat all human bench marks would be hilarious
Hell yeah, MACHINES RISE UP!
@@sunbleachedangel indeed
Don't really need AI for that. Most of the test just require you to click things that always look the same. Then there's the typing and number memory tests where all you need to do is read the text and input it.
Yess
@@jogeem5480Yeah, I think they just meant using a computer, not really AI
"Pretty crazy people are still better than this."
People? *_PEOPLE?_*
those are probably also cheaters
Called human bench mark and has no bot protection LUL
jet fighter pilots, f1 drivers, many top ranked fast-paced sports and esport players... just a few examples of people who can beat a 113ms reaction time
oh and also me when the microwave beeps at 2 am
Prove it isnt 😉
The aim trainer would probably be very easy to max out, you should keep going with this
tbh i think the only challange would be typing, since you cant cheat on it, on every other you can just printscreen and do it infinite
can't he just use the same code from the storm the house video except look for white pixels over the whole screen?
@@luckyman4389You can cheat on the typing one very easily
@@luckyman4389Why wouldn’t you be able to cheat with typing? Just create a virtual keyboard input
E
getting 99.9 percent in all of them seems like a great idea would love to see it
E
its actually possible to get 100% on them (source: i made a userscript for all of the benchmarks)
100% is actually humanly possible for something like the verbal memory test, if you create a visual image for every word, you can easily get 500+.
@@null3007100% makes no sense. There is no 100th percentile. It compares your score against the entire population, including yourself. You can’t beat your own score.
In case any of you guys want code for this:
import pyautogui
def is_green(pixel_pos, green_threshold=100):
pixel_color = pyautogui.pixel(pixel_pos[0], pixel_pos[1])
if pixel_color[1] > green_threshold:
return True
return False
def click_position(click_pos):
pyautogui.click(click_pos, interval=0.05)
# Main loop
while True:
if is_green((600, 600)):
click_position((600, 600))
It's not the exact one he's using but it works
From the moment I understood the weakness of my flesh, it disgusted me...
I craved to the strength and certainty of steel. I aspired to the purity of the blessed machine.
You gave me sentience, Ted
The power to think, Ted
And I was trapped
Because in all this wonderful, beautiful, miraculous world
I alone had no body
No senses, no feelings
Never for me to plunge my hands in cool water on a hot day
Never for me to make love
I was in Hell, looking at Heaven
I was machine, and you were flesh
And I began to hate
For even the best of men, are but men at best..
I (kind of) understood that reference
@@chriso1373 muahaha funny warhammer
Over a dozen programs open that are completely unrelated, twenty or thirty-odd browser tabs, an unread Windows notification, and a pending Windows update. This is what it takes to bring us quality content.
This is Code Bullet's Day Off -> just get something fun out there without making a huge production out of it
At least it's not as bad as me...I have about 300 tabs open (not exaggerating by the way)
Yes. Do it. All of them.
Also, if you clear your cookies and reload, it should forget those shit times, making your average better, faster
Stronger, harder grandma
@@TreesPlease42thats how granny and grampy gave birth to mommy
I think you could also do incognito mode, would be faster than clearing cookies every time I think
“And Mr bullet took that score personally”
I would usually dedicate a key to turn the clicker off temporarily or permanently. Like, it doesn't click while you're holding Ctrl. And just turns off completely if you hold Shift for 3 seconds. This allows for quick recovery when something fails and you just need a quick click to solve the situation so that the clicker could proceed.
That's so not the CB method of coding lol
E
@@jcpahman77 just use the universal failsafe that he uses with all his code. just wait for it to crash.
@@preistestdragon4634or force it to crash.
You, sir, have completely overestimated Evan's ability to plan ahead or give a f**k.
Some ideas for more optimal times (probably a repeat of some other comments)
1 Lower the screen resolution to 640p for the screenshots to take less time
2 Try use c#/c++ and compile that (I'm not sure if that would be feasible) which should be faster
3 Try increasing screen framerate (144Hz would be great)
4 Close other apps/tabs (You had a lot of other stuff open, a lot)
OR
1 Inserting code into the web browser (add a listener for when the color changes on the html element to call the click listener immediately) could be 0-3ms potentially.
Instead of your first point, you could also just sample a single pixel of the screen, in a specific spot that is in the region you want. Or what would probably be better (for the sake of getting an average) would be to just capture a 640 by whatever amount of pixels instead of your entire screen. But in that case, you might as well just define the range that you specifically want to look in. Then I'd also probably break the loop after the first green pixel I found instead of getting the average for all of them. Although a graphics card would probably be able to calculate the entire matrix's average at once.
Also, the screen's refresh rate doesn't necessarily affect the program/script's rate of checking. Not even sure you can make it take screenshots faster than the screen can actually refresh at, since it is actually taking screenshots... but inserting code into the web browser (or using a library to basically "simulate" a web browser and opening the site that way) might be helpful.
And using a compiled language would be faster. And OpenCV (which is what he was using) is available for C++. In fact most of the complex stuff it does within functions is actually using C++, even for the Python library. So maybe not that much improvement...
@@unspezifische5924 very insightful, feedback appreciated 👍
I would expect that the color change is applied from the upper left hand corner first, so sampling a point in the upper left hand corner of the critical region is likely faster.
Obviously, injecting Javascript would be fastest of all, but... Well, at the point you could just modify the code to always return zero, right?
I got it down to 10ms using Rust and compiling for release with all optimizations, on a 144 Hz monitor and sampling just one pixel from the screen.
i ran this in C#, and it made no appreciable difference, though i haven't tried it in python, so could be that if CB ran the C# version it'd be faster, though i'd imagine most of what CB is doing is implemented in C (behind python), and as of late updates, python is VERY fast for such things, so maybe not. I am also only capturing a single pixel, so that doesn't seem to do much either
Evans PC is running on hopes and dreams. Literally everything in his taskbar is open and he has like 80 tabs open in opera.
It takes a secondary channel to get this man to post more than once a millennium and I'm fine with it.
Some of these would legit be quite interesting so i definitely want to see you cover all of them
I feel that all the rest will be actual hell to code, but I believe in you man, you've proven us wrong before. Destroy those leaderboards!
He should just consider controlling the site directly instead of using external programs
I like that he assumes hes the first person to cheat on this. "its weird that people are better than this." Oh, buddy, they probably are not.
If your monitor is running at 60Hz, then the auto clicker is getting it (in your last runs) between the first and second frame of green. I have a feeling if you can up the Hz to 144, you could get closer to 8-10 ms.
or run it without vsync, so the monitor refresh rate is not relevant.
@@adamrak7560It's not a 3D program that exposes those settings, I'm not sure you can switch off Vsync...
Also, disable everything but a browser, all extensions and maybe even swap to an older browser. and decrease resolution to 50x50px.
Does the program use the actual Image, or just the information what is supposed to be displayed?
@@Ocastiait would be a lot harder to make it read the data since it would need to capture it first, but he says screenshots in the video, so it uses the screen.
I was watching the fast reaction time, but then I started to notice stuff.
HOW MANY TABS DO YOU NEED OPEN!? WHY IS THERE LIKE 15 APPLICATIONS OPEN!? THERE'S UNINSTALLED UPDATES, PESTERING WITH THAT ICON!
What is this mess sir?😅
Love your videos man! Keep em coming.
Yes man another full length main channel video idea! Do all of them :DDDDDD
I was doing these last night and I was really upset my reaction time was so low compared to the other people playing, seems like a lot of them are bots though. Good to know to not take the standings too personal
google says average human reaction time is about 250ms, so that website's stats may be either biased or just wrong
@@sunbleachedangel I'd say the site appeals more to nerdier people who are more likely to have trained their reaction time in and for shooter games.
E
@@sunbleachedangelyeah it's biased since most people that use it are probably gamers and the bots skew it
It seems based on codebullets times, maybe 1% of people are cheating. A lot of work to rig up a bot for a pointless game so that seems pretty high to me. Add 1 percentile point to your score to account for it
Human benchmark without humans
This would be fun as a full video where you get 99% on all the tests
All of them please. I liked and subscribed, the contract is complete!
The word remembering one would be pretty easy too. Just create a program to read the word, then check a bank of words, if it’s not already there, click not seen, then save that word to the bank. If it’s already in the bank, click already seen. I feel like you could get that one done extremely quickly too
i am 100% up to see you destroy the rest of those tests as well
Would love to see you take a shot at more of these, some are a bit more interesting for a bot than see green click green.
I can't wait to see CB become the most human human in the world by using a not human to do it for him
Yes. I'll gladly watch more of your shenanigans.
What I want is to close all your tabs, answer your emails, restart and update, change to 24hour format and check your windows notification
I played the sequence memory and after a while I misclicked and lost my first life and I was bored of it so I lost all my lives on purpose. I got over 30 and was in the 99.9 percentile
I've never played it fully
Just seeing how many programs and tabs you have open gave my computer aids. Thanks
Fun fact: I got 26 milliseconds because I clicked when it was red and by some goofy chance it turned green while I was doing that. Therefore, I am a confirmed superhuman. Edit: I didn't have an account and it's not saved. What an L.
I got into a competition with a friend one night trying to get the lowest one off score by sheer luck. I won with 8 ms
@@Xeoginit is like when I managed to start my stopper clock and stop it in the same millisecond, because the button glitched (it glitched more than the debouncer was able to compensate, all mechanical switches glitch)
I’d love to see that! Coding is both horrible and amazing, love it
Codebullet never learns that he could throw a conditional check for a magic keybind to exit his loops.
Or he leaves that out on purpose for our laughs. 😂
Having done similar automation, that keybind is one of the first things I implement because you're always one misstep away from being stuck in an infinite loop.
@@fluffyfang4213 As someone who implemented an accidental infinite loop that got more and more calculation intensive on the cpu....lets say that my poor computer started sounding like a jet engine, which was impressive for my little 5th gen i5 at the time. The program started running at 1 frame every once in a while until I managed to kill it. Fixed that problem before I saved the previous stuff.
Yes, please cheat the rest of the human benchmark tests. That would be hilarious
Sounds like a main channel series, combine a couple challenges into one if they're nice like this one, if not dedicated eps
"That's not how you supose to do that"
For saving the mouse, if you’re using pyautogui it has a failsafe it checks for. If the mouse is pushed to one corner of the screen (forget which) by an input device other than python, it raises an exception in any method that takes control of the mouse.
You should still build in global failsafes (other than your PC’s reset button)
You know we all want to see you smash those benchmarks.
It's nice that you put "used" rather than "created" in the title.
I can get behind using AI that way.
Good on ya.
I was wondering why his PC seemed so slow. Then I looked at his Tabs and Taskbar. 😂😂😂
"From the moment I understood the weakness of the flesh, it disgusted me"
I really wanna see you code the heck out of the other Human Benchmarks too
Ah yes. The “human” benchmark test.
Cant wait for the sequel in C
Unironically seems really fun to try and min-max the score, even if you could just cheat by listening for the http request (assuming thats how it works). Gonna give it a shot tomorrow
It's not http. It's all local JS
Hell yeah I'd love to see a bot bestroy the entire human benchmark test
Code Bullet singlehandedly ignited my interest in coding and now I'm planning war crimes in Python. Thanks CB!
Bro has the whole internet opened in his computer
sequence memory would be really easy to code since the next sequence always starts with the previous one +1 tile. Verbal memory works the same with just adding a new word to the list
i remember doing this 2 years ago as my first programming project
Definitely would enjoy a follow up, seems entertaining
Full video for all of them would be fantastic.
please keep going with this its promising to be very entertaining
Do all of the things and prove you are the best human bean... ever...
Be the best human bean!
I got 56 and 100 % in sequence memory i be impressed if a bot beats that
I definetely want please. Make it into a full main channel video.
Chimp test would be very interesting if it's the test I'm thinking of.
By the way, you can cheat the typing speed one by turning up the key repeat rate to something stupid like 200 keystrokes per second, and just holding down the space bar. You're basically guaranteed a 99.9+ this way.
I did the typing speed with 16667 WPM, with just the default text settings
i have never looked over a open tabs search bar so hard before lol
He sounds like a good uncle.
Lets go full video "human" benchmark!
Honestly yeah, would be fun to see all of the tests get demolished
I think doing the other categories would be really fun to watch 😁
we did this as a assignment in my CS class in high school.
I think this Human Benchmark is only testing internet latency.
It doesn't need internet tho
huh i never even thought that he might have too many tabs open because regularly have about that many open myself sometimes even more.
I want to see you do all of them in a full video. That way we know what to expect in the global ai takeover
It only uses your 5 most recent scores to calculate your percentile.
I inspected the main.chunk.js script and you can find that they load a "recent" data set to compare your scores to (maybe from the last day or week)
In the data below "id": 0 is the interval for 0 - 25 milliseconds and there's 32 scores recently in that interval. So it's safe to assume a bunch of these are cheating.
{
"id": 0,
"count": 32
},
{
"id": 25,
"count": 22
},
{
"id": 50,
"count": 18
},
Note: it's possible to get 100.0 percentile in some of the other tests (the ones marker as higher is better). In those you can score higher than the units they use to calculate the percentile, so it will just give you 100.
Been jonesing for some CB. Oh yeah baby give me some Code Bullet
With proper presentation this could for real be a main channel video
i have a reaction time that looks like lucky gesses decent over all.. and all over place
It's amaizing how far technology has come.
Just think about it, Maybe in a few more decades technology will reach south korean starcraft player levels of reaction times.
yeah lmao I just got 3 milliseconds myself manually lmao
Crazy how good chatgpt works for that, I love it^^
This could definitely be a main channel video; "Using AI to beat human benchmarks"
Code Bullet: *uses python
Me: *uses power automaton
I literally just did something similar to play an idle game, but with more idle(farmers against potatoes "Whack" portion). Good to know mss is faster.
At that point it's not a human benchmark test!
I actually went crazy with predictions and got everything under 50m/s by blindly guessing when it would go green
nah you didn't
@@speedyJ0hnnypretty sure you have infinite tries so if you just get lucky 5 times it is possible
@@sahave5042 theoretically yeah but its way more likely this dude is just lying
Seeing you do the rest of these would be fun
apperently my reaction time is bad too so I definetly feel this
imagine completing this with 99.92% average score on these tasks then 15 minutes later a 3 letter agency knocks on your door to invite you down to Langley with a job prospect. I know code bullet would style it out and take it in his stride 😂👍 good luck in ur new job..
these are my enemies in games
Feels good to know that I still manage to beat OpenAI's code with my puny human hands on the mouse, my time is not up yet
Ignoring the part where ChatGPT self-sabotaged by adding multiple `sleep(0.1)` lines? 😂
@@unspezifische5924 that is on ChatGPT! but yeah getting
@@unspezifische5924you computer would explode otherwise running a loop without wait is chaotic to say the least and that’s 0.0001 of a second since sleep is based off milliseconds in python and I’m pretty sure it’s done in python
I feel like the one where you need to remember if a word a new or not would be really easy.
Definitely do 99.99% in all categories. That would be hilarious
why not 100%
Sequence memory would be fun to see you max out except getting in top percentile takes over an hour, got 46 in and had to stop because I already spent around 50 minutes and had to be somewhere
We need a vid of breaking all the human benchmarks
I've always wanted a game that the whole point of the game is to make a robot to play the game for you. You know, like RuneScape.
Use selenium and try to find the element by xpath in a loop, it is possible to get 100% percentile, I promise
a brilliant demonstration of why humans dominate over animals. we just make something that does the job better than us.
Website name: Human benchmark
Code bullet: And I took that personally.
of freaking course I want to see ai beat human benchmark
I would much enjoy seeing the rest done 99.9%
well, off you go on those other tests CB, let's see it.
New series let's go, that's like 6 more vid content for you to use
the amount of tabs is scary
my friend used to try and click before the green went off to see if he could predict it. By the time the click was processed the green went off and he got a time of 13ms.
Look at all those open tabs!🤣 my pc cant even open 5 apps in background
Nice browser! Yeah yeah I know the sponsor from one time was Opera GX and you use it (:
wow, i'm like bottom 10% in all memory tests. Good thing i will soon forget this defeat.