tuco wrote:I will repeat, I understand that crank. Auto exposure metering, for example, uses algorithm and can also be fed data for patter recognition. I have already said that AI can be fed much more data and much faster than any human. Period.

What good is more data if you can't work with the data that you already have? What don't you understand about we just aren't there yet. Do you think machines can do pattern recognition better than humans? Can understand speech for example, better than humans? Just look at captcha for another example.

Are you trolling me?

I have said numerous times "in principle" - the ultimate jail out card - which means nothing else than "its possible" - unlike impossible - and on top of it I have said:

tuco wrote: So when informed that the tech is not available yet I said:

and you tell me, again, I do not understand? You have no idea how annoying this kind of discourse is to me. If this is not trolling, I dunno what is.

tuco wrote:

Weaver wrote:

tuco wrote:You explained nothing. You posted cool story about unconsciousness and how special it is, sauce. I am not idiot, rather the opposite , I understand that. Humans walk effortlessly, unlike robots. So what? Sniping is "mechanical" activity, scientifically quantifiable, putting variables into formula. Period.

Prove it. Quantify the variables in the wind videos I posted - provide the formulas that will yield wind speed and direction at the specified distances from the camera.

Honestly I am losing my patience but that is my cool story. Its like asking to provide formulas for water drop distribution when an excrement hits water in toilet. In principle, such distribution can be accounted for precisely. Beyond me why and how would someone dispute this.

In principal, many things are possible. That doesn't mean they will ever be practically. I doubt your example would, the needed resolution of the 3D shape of the turd would need to be extreme, and you'd probably need to know the velocity of most of the molecules involved. There is a lot of microspray going on, a sort of fog of santorum you have to account for to do a thorough job. I think a machine that can shoot far more accurately than a human is possible, but we're not very close yet.

You doubt? Well, you know - because you recognize patterns - what will follow now. Cool story, bro. I do not doubt it. Now what?

---
edit: What you and others here do not understand is that this is exactly why debating cool stories/opinions is essentially pointless. I think, you think, she thinks and whoever else thinks .. and the only way to tell who is closer to reality - apparently because I do not agree with such assessment - is how s/he told the cool story, whether it makes sense or seems reasonable or believable or credible or even how craft-fully worded it is, which is totally ridiculous. Philosophy 101 or rather science 101.

edit1: and since nobody ever fucking apologizes here I will do it myself

Oh I missed that, sorry mate.
No problem dude, happens.

tuco wrote:I have said numerous times "in principle"

tuco wrote:I have already said that AI can be fed much more data and much faster than any human.

In principle, there is so much data that it would take a computer too much time to do the computation, even given that information travels around the computer at the speed of light. Remember, the shot we are talking about unfolded in ten seconds or less. This is called computational complexity, and is what AI has to deal with, since it is artificial and not 'genuine'. If the computer takes too much time, the target moves, and the computation has to begin again, not least because (as has received focus in this problem) it is about the air in the space between the muzzle and the target. I've got it! Put a guidance computer and a vision system inside the bullet! "In principle", it would fit!

Reassure us that the amount of data necessary to solve this problem is therefore limited to what a computer can solve in a few seconds. Are we discretizing the trajectory, or just winging it? Or perhaps, in principle, information, can travel faster than c, but that's the 'cartoon' principle or the 'evil clown' principle. Remember, if the bullet goes fast enough in air, it melts. Phasers on 'distintegrate', lads. Prepare to abolish the inverse square laws!

Wanted to say touché then you said "has to begin again". In principle and for FTL we've had other threads but I understand that the sauce around a point is hallmark of your communication here.

That is why I asked to list the variables needed. Then we could go over them one by one and see, or rather guess, whether or not there are physical limits similar to solving chess problem. Tho even there the jury is out .. not in principle but in fact. I got wind at target. Well, perhaps I am indeed ignorant fuck with wilder than wild imagination but it does not seem to me like too much data.

tuco wrote:Wanted to say touché then you said "has to begin again". In principle and for FTL we've had other threads but I understand that the sauce around a point is hallmark of your communication here.

That is why I asked to list the variables needed. Then we could go over them one by one and see, or rather guess, whether or not there are physical limits similar to solving chess problem. Tho even there the jury is out .. not in principle but in fact. I got wind at target. Well, perhaps I am indeed ignorant fuck with wilder than wild imagination but it does not seem to me like too much data.

I listed the air in the space between the muzzle and the target, and how precisely you have to know about it to control the trajectory to end up in an appropriately-sized circle at the target. It's already been alluded that luck was involved in achieving the final circle or point at the target. If you give the problem over to AI, you're bound to end up with a probability distribution for the end point of the trajectory, because the air is moving. This was mentioned in the early stages of the thread, and I don't know what aspect of AI you are envisioning to account for this and get beyond a probability distribution. Remember, we are shooting to kill.

Luck is same for human and AI ;) So again, what is the special sauce humans have that they can estimate this air in the space variable and AI cannot?

tuco wrote:Luck is same for human and AI So again, what is the special sauce humans have that they can estimate this air in the space?

They don't use the vast quantities of data you were only recently envisioning feeding to your AI sniper.

So, I think that's a good place to stop, because I thought you were implying that AI could do better. Maybe you weren't, but you didn't mention any of the points I brought up.

Tuco, what's your longest shot at say, a beer can sized target?


Sent from my iPad using Tapatalk

Wanna hear cool story, really?

When in high school it was mandatory for boys to take part in shooting at target from something like this:



The best shooters would then be approached by military and offered *coughs* career in military. Defending against the evil imperialists and stuff. So me and my buddy Georgios, whom with I went to see the Platoon just weeks earlier, focused on recreating the shooting scenes from the movie. Unfortunately, we liked most where they shot automatic weapons and empty carriages were flying all around tho it did not stop up and we were reloading like mad emptying the magazine as quick as possible. I think Georgios got 1 out of 10 (1 or 2 points) and iirc I got 0 out of 10. So we were not selected but the biggest idiot of our class Vaněk was and he even went to study military degree or something.

Longest shot? I dunno but I could hit paper roses and Hubba Bubba chewing gums alright



So what is this? You are not expert, scholar, so you dunno what you are talking about? Is that it?

Cito di Pense wrote:

tuco wrote:Luck is same for human and AI ;) So again, what is the special sauce humans have that they can estimate this air in the space?

They don't use the vast quantities of data you were only recently envisioning feeding to your AI sniper.

So, I think that's a good place to stop, because I thought you were implying that AI could do better. Maybe you weren't, but you didn't mention any of the points I brought up.

I was envisioning the AI could learn from vast amount of data.

Waddya think, Tuco? Not knowing the problems, you think maybe you're talking out of your ass?

My longest routine range with a 6mm Remington model 700 on a bipod with 12x optics is around 350 meters. I can hit that beer can sized target at that range just about every time.

And yes, that's exactly it. You have no concepts of the problems in making that shot. With that lack of understanding, your assessment of the suitability of "AI" to replace snipers is baseless.


Sent from my iPad using Tapatalk

Now, what I didn't read in all this was how many shots this guy missed at that range.


Sent from my iPad using Tapatalk

The concept is straightforward. There are scientifically quantifiable variables and there are laws of physics. There is no mystery, there is no special sauce. The only thing there is is challenge for AI researchers to come up with appropriate solutions, code monkeys code them and robots manufacture another robot.

I mean, how does a sniper estimate "the air in the space" factor? Years of experience and training plus skill. Well, yes that is probably how I would explain it to a child because I would not know any better.

The_Metatron wrote:Now, what I didn't read in all this was how many shots this guy missed at that range.


Sent from my iPad using Tapatalk

I asked the same question a couple of times. It most definitely is a matter of luck, the shooter can't know all the information needed, and the wind is not constant, so luck. Do these guys train or practice for such distances? I know a lot of BMG fans take ridiculously long shots trying to hit large boulders and such, it's been a long time but I think they're going for multiple miles. I had one back in 2001, I didn't go in for distance, I was more interested in destruction. I grabbed this Yugo that was red-tagged as abandoned and had a lot of fun with it. The animated gif was from an incendiary round. Sorta became more of an Itwent at that point.

Damn, I'd love to shoot one of those bastards.

tuco wrote:The concept is straightforward. There are scientifically quantifiable variables and there are laws of physics. There is no mystery, there is no special sauce. The only thing there is is challenge for AI researchers to come up with appropriate solutions, code monkeys code them and robots manufacture another robot.

I mean, how does a sniper estimate "the air in the space" factor? Years of experience and training plus skill. Well, yes that is probably how I would explain it to a child because I would not know any better.

Sure. That's all.


Sent from my iPad using Tapatalk

tuco wrote:The concept is straightforward. There are scientifically quantifiable variables and there are laws of physics. There is no mystery, there is no special sauce. The only thing there is is challenge for AI researchers to come up with appropriate solutions, code monkeys code them and robots manufacture another robot.

The variable that's been put on the table is wind speed and direction along the bullet's trajectory. This is something you should be easily able to measure and get large quantities of data on. Now what are you going to do with that data? The problem you are being asked to solve is: given a video feed, predict wind speed and direction along the bullet's trajectory. So far, only the output here is easily quantifiable data. The input is not easily quantifiable data. The raw variables in the input comes in the form of a huge stream of binary data.

How do we get a function from that data stream to the quantifiable output?

I still have questions about the laser approach. Dumb question, but can't you point the laser slightly away from the target, so they're not likely to see it, and take enough samples that way to estimate wind speed and direction in the intended trajectory?

VazScep wrote:

tuco wrote:The concept is straightforward. There are scientifically quantifiable variables and there are laws of physics. There is no mystery, there is no special sauce. The only thing there is is challenge for AI researchers to come up with appropriate solutions, code monkeys code them and robots manufacture another robot.

The variable that's been put on the table is wind speed and direction along the bullet's trajectory. This is something you should be easily able to measure and get large quantities of data on. Now what are you going to do with that data? The problem you are being asked to solve is: given a video feed, predict wind speed and direction along the bullet's trajectory. So far, only the output here is easily quantifiable data. The input is not easily quantifiable data. The raw variables in the input comes in the form of a huge stream of binary data.

How do we get a function from that data stream to the quantifiable output?

I still have questions about the laser approach. Dumb question, but can't you point the laser slightly away from the target, so they're not likely to see it, and take enough samples that way to estimate wind speed and direction in the intended trajectory?

On a distance scale of 3 km, isn't the atmosphere (even a dry one) convecting turbulently over ground that has sun shining on it? Best to take your shot before the sun is well up. The point isn't even to get AI to do a job just as well as a human can, but to process 'vast quantities of data' and find solutions that humans cannot. This isn't about finding imbalances to exploit in equities prices. Anyway, sniper kills are as morally questionable as exploiting price imbalances.

It's a pure mistake to conclude that because there isn't any special sauce, that there are not limits to achievement in some field of endeavour. When this guy hit a target at distance 3+ km, it was as if Usain Bolt is running 100 m in 9 seconds and somebody comes along and runs it in 6 seconds. Excuse the analogy between athletic strength and shooting skill. Until more data appear, it looks like an outlier, and someone has already asked how many misses there were before that kill.