What is really going on at KTM?

3strokemx
Posts
2520
Joined
9/2/2010
Location
US
10/15/2025 10:53am

Once you guys lose your virginity I think the novelty of this AI nonsense will wear off.

29
3
mxaniac
Posts
556
Joined
9/9/2019
Location
Airway Heights, WA US
10/15/2025 11:53am
3strokemx wrote:

Once you guys lose your virginity I think the novelty of this AI nonsense will wear off.

It's a useful tool, people are just starting to figure out what it's good for.

It works on probability, and has access to boatloads of information. It will never be good for sensitive topics like social issues because of all the intervention. On the other hand, if I give it a decent amount of information, so that everything isn't assumed based on probability, it can write a pretty good technical proposal saving me lots of time. It can also write simple scripts, strip this out of these 500 documents. I've never successfully had Claude, ChatGPT, or Grok program anything complex though. They guess at function names, variable types, and API calls based on probability instead of actually looking things up. AI is also great for finding research data on specific topics. The neural network based AI is also really good for things like defect detection. It used to be you had to train AI on what a defect was, Every type of defect that could ever occur. Now you train it on what good looks like, and then the infinite other possibilities are defects. Much better!

So how would this relate to MX? I would say correlating telemetry data with lap times to optimize ecu settings and suspension setup for one thing. Evaluation of accident reports to help optimize protective gear for the most typical injury prevention. What won't it be good for? Predicting the unpredictable which race outcomes are in general. Brand comparisons because the data it processes is all marketing drivel. Determining a shock failure, because it has no real pertinent information to go by. All it can do is predict based on past failures.

I hated Nirvana and Drew Bledsoe for a while, because at the time that's all I heard about. Nothing wrong with either, I was the problem for not just being patient.

3
9
Axlnut_KM3
Posts
117
Joined
11/20/2024
Location
EAST WATERFORD, PA US
10/15/2025 12:23pm
3strokemx wrote:

Once you guys lose your virginity I think the novelty of this AI nonsense will wear off.

Yeah, like the dang kids and their dang interweb computators. 

2
6
Jkawi
Posts
504
Joined
3/5/2015
Location
CA
10/15/2025 12:31pm
MotoDad32 wrote:
I'd guess it's all relative.  When the shock is ~300 degrees, that ~200 degree air coming off the radiator has a cooling effect.  I'm making up...

I'd guess it's all relative.  When the shock is ~300 degrees, that ~200 degree air coming off the radiator has a cooling effect.  I'm making up those numbers, but that's the general gist.

Jabe wrote:
It’s the basic principle of heat transfer. It’s all about the delta and not the absolute temperature values.Think about the cooling of turbine blades in a...

It’s the basic principle of heat transfer. It’s all about the delta and not the absolute temperature values.

Think about the cooling of turbine blades in a jet engine. The absolute temperature of the cooling gas flow is ridiculously high. But still lower than the temperature in the turbine.

Which makes you wonder why they went to the extra effort to make it suck from behind the rad. Just duct it to open air...

1

The Shop

3strokemx
Posts
2520
Joined
9/2/2010
Location
US
10/15/2025 12:43pm
3strokemx wrote:

Once you guys lose your virginity I think the novelty of this AI nonsense will wear off.

mxaniac wrote:
It's a useful tool, people are just starting to figure out what it's good for.It works on probability, and has access to boatloads of information. It...

It's a useful tool, people are just starting to figure out what it's good for.

It works on probability, and has access to boatloads of information. It will never be good for sensitive topics like social issues because of all the intervention. On the other hand, if I give it a decent amount of information, so that everything isn't assumed based on probability, it can write a pretty good technical proposal saving me lots of time. It can also write simple scripts, strip this out of these 500 documents. I've never successfully had Claude, ChatGPT, or Grok program anything complex though. They guess at function names, variable types, and API calls based on probability instead of actually looking things up. AI is also great for finding research data on specific topics. The neural network based AI is also really good for things like defect detection. It used to be you had to train AI on what a defect was, Every type of defect that could ever occur. Now you train it on what good looks like, and then the infinite other possibilities are defects. Much better!

So how would this relate to MX? I would say correlating telemetry data with lap times to optimize ecu settings and suspension setup for one thing. Evaluation of accident reports to help optimize protective gear for the most typical injury prevention. What won't it be good for? Predicting the unpredictable which race outcomes are in general. Brand comparisons because the data it processes is all marketing drivel. Determining a shock failure, because it has no real pertinent information to go by. All it can do is predict based on past failures.

I hated Nirvana and Drew Bledsoe for a while, because at the time that's all I heard about. Nothing wrong with either, I was the problem for not just being patient.

AI for tasks is fine, should make life easier. 

AI for analysis is classic confirmation bias.  

It's like the example of WW2 airplanes returning with bullet holes.  AI would tell you where every single bullet hole is, so you can reinforce where the planes get shot.

A smart person would tell you to look at where the returning planes had been shot, and reinforce different areas, because these are the planes that made it back.

18
mxaniac
Posts
556
Joined
9/9/2019
Location
Airway Heights, WA US
10/15/2025 1:09pm
3strokemx wrote:

Once you guys lose your virginity I think the novelty of this AI nonsense will wear off.

mxaniac wrote:
It's a useful tool, people are just starting to figure out what it's good for.It works on probability, and has access to boatloads of information. It...

It's a useful tool, people are just starting to figure out what it's good for.

It works on probability, and has access to boatloads of information. It will never be good for sensitive topics like social issues because of all the intervention. On the other hand, if I give it a decent amount of information, so that everything isn't assumed based on probability, it can write a pretty good technical proposal saving me lots of time. It can also write simple scripts, strip this out of these 500 documents. I've never successfully had Claude, ChatGPT, or Grok program anything complex though. They guess at function names, variable types, and API calls based on probability instead of actually looking things up. AI is also great for finding research data on specific topics. The neural network based AI is also really good for things like defect detection. It used to be you had to train AI on what a defect was, Every type of defect that could ever occur. Now you train it on what good looks like, and then the infinite other possibilities are defects. Much better!

So how would this relate to MX? I would say correlating telemetry data with lap times to optimize ecu settings and suspension setup for one thing. Evaluation of accident reports to help optimize protective gear for the most typical injury prevention. What won't it be good for? Predicting the unpredictable which race outcomes are in general. Brand comparisons because the data it processes is all marketing drivel. Determining a shock failure, because it has no real pertinent information to go by. All it can do is predict based on past failures.

I hated Nirvana and Drew Bledsoe for a while, because at the time that's all I heard about. Nothing wrong with either, I was the problem for not just being patient.

3strokemx wrote:
AI for tasks is fine, should make life easier. AI for analysis is classic confirmation bias.  It's like the example of WW2 airplanes returning with bullet holes...

AI for tasks is fine, should make life easier. 

AI for analysis is classic confirmation bias.  

It's like the example of WW2 airplanes returning with bullet holes.  AI would tell you where every single bullet hole is, so you can reinforce where the planes get shot.

A smart person would tell you to look at where the returning planes had been shot, and reinforce different areas, because these are the planes that made it back.

I agree, AI would be great at taking the bullet hole data and giving you a Pareto.  You would then need to use judgment and logic to interpret what to do about it. Would AI know what to do with the data? Of course yes in this particular instance, because this is a known "problem" and if you ask AI the question it knows all about Abraham Wald and his study. The question is, can AI correlate that to a different scenario and I doubt it.

As for what it's going on at KTM, I don't see how AI can assist with their financial trouble, supply chain trouble, plant relocation, or conflicting agendas between them and Bajaj. It would likely be great at finding historical analogs though.

coopernicus
Posts
290
Joined
12/15/2019
Location
Broomfield, CO US
10/15/2025 1:38pm
MotoDad32 wrote:
I'd guess it's all relative.  When the shock is ~300 degrees, that ~200 degree air coming off the radiator has a cooling effect.  I'm making up...

I'd guess it's all relative.  When the shock is ~300 degrees, that ~200 degree air coming off the radiator has a cooling effect.  I'm making up those numbers, but that's the general gist.

Jabe wrote:
It’s the basic principle of heat transfer. It’s all about the delta and not the absolute temperature values.Think about the cooling of turbine blades in a...

It’s the basic principle of heat transfer. It’s all about the delta and not the absolute temperature values.

Think about the cooling of turbine blades in a jet engine. The absolute temperature of the cooling gas flow is ridiculously high. But still lower than the temperature in the turbine.

Jkawi wrote:

Which makes you wonder why they went to the extra effort to make it suck from behind the rad. Just duct it to open air...

Opinion from a retired engineer: They would duct warmer air to keep the shock temperature in a more narrow range.  In that case, they would not want the shock to get "too cool". It's possible the damping with a cooler shock affects the "comfort" (sorry to use that term) the rider desires and all this time we thought fading was the problem it could have been the shock getting too cool and too stiff! 

5
5
Axlnut_KM3
Posts
117
Joined
11/20/2024
Location
EAST WATERFORD, PA US
10/16/2025 8:36am
mxaniac wrote:
It's a useful tool, people are just starting to figure out what it's good for.It works on probability, and has access to boatloads of information. It...

It's a useful tool, people are just starting to figure out what it's good for.

It works on probability, and has access to boatloads of information. It will never be good for sensitive topics like social issues because of all the intervention. On the other hand, if I give it a decent amount of information, so that everything isn't assumed based on probability, it can write a pretty good technical proposal saving me lots of time. It can also write simple scripts, strip this out of these 500 documents. I've never successfully had Claude, ChatGPT, or Grok program anything complex though. They guess at function names, variable types, and API calls based on probability instead of actually looking things up. AI is also great for finding research data on specific topics. The neural network based AI is also really good for things like defect detection. It used to be you had to train AI on what a defect was, Every type of defect that could ever occur. Now you train it on what good looks like, and then the infinite other possibilities are defects. Much better!

So how would this relate to MX? I would say correlating telemetry data with lap times to optimize ecu settings and suspension setup for one thing. Evaluation of accident reports to help optimize protective gear for the most typical injury prevention. What won't it be good for? Predicting the unpredictable which race outcomes are in general. Brand comparisons because the data it processes is all marketing drivel. Determining a shock failure, because it has no real pertinent information to go by. All it can do is predict based on past failures.

I hated Nirvana and Drew Bledsoe for a while, because at the time that's all I heard about. Nothing wrong with either, I was the problem for not just being patient.

3strokemx wrote:
AI for tasks is fine, should make life easier. AI for analysis is classic confirmation bias.  It's like the example of WW2 airplanes returning with bullet holes...

AI for tasks is fine, should make life easier. 

AI for analysis is classic confirmation bias.  

It's like the example of WW2 airplanes returning with bullet holes.  AI would tell you where every single bullet hole is, so you can reinforce where the planes get shot.

A smart person would tell you to look at where the returning planes had been shot, and reinforce different areas, because these are the planes that made it back.

mxaniac wrote:
I agree, AI would be great at taking the bullet hole data and giving you a Pareto.  You would then need to use judgment and logic...

I agree, AI would be great at taking the bullet hole data and giving you a Pareto.  You would then need to use judgment and logic to interpret what to do about it. Would AI know what to do with the data? Of course yes in this particular instance, because this is a known "problem" and if you ask AI the question it knows all about Abraham Wald and his study. The question is, can AI correlate that to a different scenario and I doubt it.

As for what it's going on at KTM, I don't see how AI can assist with their financial trouble, supply chain trouble, plant relocation, or conflicting agendas between them and Bajaj. It would likely be great at finding historical analogs though.

You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story and said "aha, makes sense"

Instead, it was one group (SRG, at Columbia University) and one prominent member of that group (Abraham Wald) who came up with the theory of reinforcing the other areas. It was groundbreaking in the operational research field, was not widely accepted at first, and is used a prominent example in survivorship bias and statistical engineering to this day.

It's actually a better example of how we aren't really all that smart as individuals, and are very susceptible to bias confirmation and institutional and theoretical inertia even in technical fields. Machine learning is helping to break that and think outside the box.

Even our very basic (today) GPT5 and similar could lead you down the same road with the right team asking it the right questions. Just much faster than the team could on it's own - and it gets better every day.

The idea that AI isn't being used to say, make KTM more profitable, or in the engineering fixes / production changes to it's bikes is almost laughable, and it will only become more prominent in that role.

This genie isn't going back in the bottle.

3
2
3strokemx
Posts
2520
Joined
9/2/2010
Location
US
10/16/2025 8:58am
3strokemx wrote:
AI for tasks is fine, should make life easier. AI for analysis is classic confirmation bias.  It's like the example of WW2 airplanes returning with bullet holes...

AI for tasks is fine, should make life easier. 

AI for analysis is classic confirmation bias.  

It's like the example of WW2 airplanes returning with bullet holes.  AI would tell you where every single bullet hole is, so you can reinforce where the planes get shot.

A smart person would tell you to look at where the returning planes had been shot, and reinforce different areas, because these are the planes that made it back.

mxaniac wrote:
I agree, AI would be great at taking the bullet hole data and giving you a Pareto.  You would then need to use judgment and logic...

I agree, AI would be great at taking the bullet hole data and giving you a Pareto.  You would then need to use judgment and logic to interpret what to do about it. Would AI know what to do with the data? Of course yes in this particular instance, because this is a known "problem" and if you ask AI the question it knows all about Abraham Wald and his study. The question is, can AI correlate that to a different scenario and I doubt it.

As for what it's going on at KTM, I don't see how AI can assist with their financial trouble, supply chain trouble, plant relocation, or conflicting agendas between them and Bajaj. It would likely be great at finding historical analogs though.

Axlnut_KM3 wrote:
You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story...

You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story and said "aha, makes sense"

Instead, it was one group (SRG, at Columbia University) and one prominent member of that group (Abraham Wald) who came up with the theory of reinforcing the other areas. It was groundbreaking in the operational research field, was not widely accepted at first, and is used a prominent example in survivorship bias and statistical engineering to this day.

It's actually a better example of how we aren't really all that smart as individuals, and are very susceptible to bias confirmation and institutional and theoretical inertia even in technical fields. Machine learning is helping to break that and think outside the box.

Even our very basic (today) GPT5 and similar could lead you down the same road with the right team asking it the right questions. Just much faster than the team could on it's own - and it gets better every day.

The idea that AI isn't being used to say, make KTM more profitable, or in the engineering fixes / production changes to it's bikes is almost laughable, and it will only become more prominent in that role.

This genie isn't going back in the bottle.

How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside the box)? 

Why does AI need "the right team asking the right questions" to come to the best resolution?  What's the limiting factor?

2
3
Broseph
Posts
1196
Joined
4/28/2018
Location
Stevenson, WA US
10/16/2025 9:18am

All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype. We did get internet on the refrigerator though.


AI is great at compiling data. Wake me up when it has an original thought. 

8
3
Axlnut_KM3
Posts
117
Joined
11/20/2024
Location
EAST WATERFORD, PA US
10/16/2025 9:22am
mxaniac wrote:
I agree, AI would be great at taking the bullet hole data and giving you a Pareto.  You would then need to use judgment and logic...

I agree, AI would be great at taking the bullet hole data and giving you a Pareto.  You would then need to use judgment and logic to interpret what to do about it. Would AI know what to do with the data? Of course yes in this particular instance, because this is a known "problem" and if you ask AI the question it knows all about Abraham Wald and his study. The question is, can AI correlate that to a different scenario and I doubt it.

As for what it's going on at KTM, I don't see how AI can assist with their financial trouble, supply chain trouble, plant relocation, or conflicting agendas between them and Bajaj. It would likely be great at finding historical analogs though.

Axlnut_KM3 wrote:
You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story...

You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story and said "aha, makes sense"

Instead, it was one group (SRG, at Columbia University) and one prominent member of that group (Abraham Wald) who came up with the theory of reinforcing the other areas. It was groundbreaking in the operational research field, was not widely accepted at first, and is used a prominent example in survivorship bias and statistical engineering to this day.

It's actually a better example of how we aren't really all that smart as individuals, and are very susceptible to bias confirmation and institutional and theoretical inertia even in technical fields. Machine learning is helping to break that and think outside the box.

Even our very basic (today) GPT5 and similar could lead you down the same road with the right team asking it the right questions. Just much faster than the team could on it's own - and it gets better every day.

The idea that AI isn't being used to say, make KTM more profitable, or in the engineering fixes / production changes to it's bikes is almost laughable, and it will only become more prominent in that role.

This genie isn't going back in the bottle.

3strokemx wrote:
How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside...

How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside the box)? 

Why does AI need "the right team asking the right questions" to come to the best resolution?  What's the limiting factor?

At least this is an honest question.

An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but the size of your repository (theoretically, we don't really know) but certainly the speed at which you can access it, add to it, or cross reference it is infinitely smaller than even our infant LLMs we are seeing now on the public side.  Of course it's data is from known data. 99.99% of all data, recollection, research etc that humans use is from known data - what else is there?

You, nor I, nor anyone can truly think "outside the box" we just reference what we know, with some probabilities, and sometimes some guesses (based on something else we know). LLMs do the exact same thing, faster, with larger data sets. The reason you need the right team is the same reason you need the right team for ANYTHING, irrespective of AI.

I can't get a trash collector to use AI to help solve a particularly daunting structural engineering task, because he doesn't even know the basics. I can get AI to help run potential solution after solution for an experienced structural engineering team, much faster than they can on their own, each to it's logical conclusion, thus increasing the efficiency of the team, and they are there to check AIs work just like they would their own. The project now took 1 year and 20 million dollars instead of 4 years and 75 million dollars. 

This is real world, happening every day. If your only experience is people making memes and asking it stupid questions about nuanced topics, yeah, it seems dumb. That's a user based problem. I use it at work every day. There's quite a few engineers who now don't get my calls or emails because AI is faster than them, more polite, friendlier, works harder, owns it's mistakes, etc. 

 

4
2
Axlnut_KM3
Posts
117
Joined
11/20/2024
Location
EAST WATERFORD, PA US
10/16/2025 9:31am
Axlnut_KM3 wrote:
You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story...

You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story and said "aha, makes sense"

Instead, it was one group (SRG, at Columbia University) and one prominent member of that group (Abraham Wald) who came up with the theory of reinforcing the other areas. It was groundbreaking in the operational research field, was not widely accepted at first, and is used a prominent example in survivorship bias and statistical engineering to this day.

It's actually a better example of how we aren't really all that smart as individuals, and are very susceptible to bias confirmation and institutional and theoretical inertia even in technical fields. Machine learning is helping to break that and think outside the box.

Even our very basic (today) GPT5 and similar could lead you down the same road with the right team asking it the right questions. Just much faster than the team could on it's own - and it gets better every day.

The idea that AI isn't being used to say, make KTM more profitable, or in the engineering fixes / production changes to it's bikes is almost laughable, and it will only become more prominent in that role.

This genie isn't going back in the bottle.

3strokemx wrote:
How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside...

How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside the box)? 

Why does AI need "the right team asking the right questions" to come to the best resolution?  What's the limiting factor?

Axlnut_KM3 wrote:
At least this is an honest question.An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but...

At least this is an honest question.

An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but the size of your repository (theoretically, we don't really know) but certainly the speed at which you can access it, add to it, or cross reference it is infinitely smaller than even our infant LLMs we are seeing now on the public side.  Of course it's data is from known data. 99.99% of all data, recollection, research etc that humans use is from known data - what else is there?

You, nor I, nor anyone can truly think "outside the box" we just reference what we know, with some probabilities, and sometimes some guesses (based on something else we know). LLMs do the exact same thing, faster, with larger data sets. The reason you need the right team is the same reason you need the right team for ANYTHING, irrespective of AI.

I can't get a trash collector to use AI to help solve a particularly daunting structural engineering task, because he doesn't even know the basics. I can get AI to help run potential solution after solution for an experienced structural engineering team, much faster than they can on their own, each to it's logical conclusion, thus increasing the efficiency of the team, and they are there to check AIs work just like they would their own. The project now took 1 year and 20 million dollars instead of 4 years and 75 million dollars. 

This is real world, happening every day. If your only experience is people making memes and asking it stupid questions about nuanced topics, yeah, it seems dumb. That's a user based problem. I use it at work every day. There's quite a few engineers who now don't get my calls or emails because AI is faster than them, more polite, friendlier, works harder, owns it's mistakes, etc. 

 

Going to add - when you sit a human team of engineers down, to say, solve the shock problem, and say "no stupid questions, no stupid answers, all ideas get the whiteboard and discussion"

You may mean that. 

Tom gives his stupid idea - with a brilliant team, maybe Tom's idea has one glimmer that sparks something else and solutions are found. That's a very effective perfect case example. Everyone has to have some level of agreeability for this to function.

Engineers don't tend to be agreeable. Tom can't help it. He loved his idea, when you critiqued part of it, trying to be polite (but you know, getting late, kids need to get to soccer practice, wife has been bugging you all day about a bill, etc) you were less than polite. 

Tom is now slightly hurt, Tom shares less. A brilliant solution goes undiscussed.

If AI is on the team, it's always polite to Tom. It's not emotionally invested, nor does it have somewhere else to be. If it presents an idea, it doesn't care if you literally call it retarded, it just presents it's next idea, It's ideas come from the entire data set it was trained on, potentially all of engineering ever written down. 

I don't want Tom gone. I just want the AI keeping Tom and I cruising at Mach1 with an endless flow of ideas, a buffer, no feelings, etc. 

6
Tim507
Posts
3477
Joined
6/8/2010
Location
Oregon City, OR US
10/16/2025 9:39am
3strokemx wrote:
How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside...

How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside the box)? 

Why does AI need "the right team asking the right questions" to come to the best resolution?  What's the limiting factor?

Axlnut_KM3 wrote:
At least this is an honest question.An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but...

At least this is an honest question.

An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but the size of your repository (theoretically, we don't really know) but certainly the speed at which you can access it, add to it, or cross reference it is infinitely smaller than even our infant LLMs we are seeing now on the public side.  Of course it's data is from known data. 99.99% of all data, recollection, research etc that humans use is from known data - what else is there?

You, nor I, nor anyone can truly think "outside the box" we just reference what we know, with some probabilities, and sometimes some guesses (based on something else we know). LLMs do the exact same thing, faster, with larger data sets. The reason you need the right team is the same reason you need the right team for ANYTHING, irrespective of AI.

I can't get a trash collector to use AI to help solve a particularly daunting structural engineering task, because he doesn't even know the basics. I can get AI to help run potential solution after solution for an experienced structural engineering team, much faster than they can on their own, each to it's logical conclusion, thus increasing the efficiency of the team, and they are there to check AIs work just like they would their own. The project now took 1 year and 20 million dollars instead of 4 years and 75 million dollars. 

This is real world, happening every day. If your only experience is people making memes and asking it stupid questions about nuanced topics, yeah, it seems dumb. That's a user based problem. I use it at work every day. There's quite a few engineers who now don't get my calls or emails because AI is faster than them, more polite, friendlier, works harder, owns it's mistakes, etc. 

 

Axlnut_KM3 wrote:
Going to add - when you sit a human team of engineers down, to say, solve the shock problem, and say "no stupid questions, no stupid...

Going to add - when you sit a human team of engineers down, to say, solve the shock problem, and say "no stupid questions, no stupid answers, all ideas get the whiteboard and discussion"

You may mean that. 

Tom gives his stupid idea - with a brilliant team, maybe Tom's idea has one glimmer that sparks something else and solutions are found. That's a very effective perfect case example. Everyone has to have some level of agreeability for this to function.

Engineers don't tend to be agreeable. Tom can't help it. He loved his idea, when you critiqued part of it, trying to be polite (but you know, getting late, kids need to get to soccer practice, wife has been bugging you all day about a bill, etc) you were less than polite. 

Tom is now slightly hurt, Tom shares less. A brilliant solution goes undiscussed.

If AI is on the team, it's always polite to Tom. It's not emotionally invested, nor does it have somewhere else to be. If it presents an idea, it doesn't care if you literally call it retarded, it just presents it's next idea, It's ideas come from the entire data set it was trained on, potentially all of engineering ever written down. 

I don't want Tom gone. I just want the AI keeping Tom and I cruising at Mach1 with an endless flow of ideas, a buffer, no feelings, etc. 

Then there is the mirror aspect of AI!  That is an absolutely amazing path to go down. I often refer to the path as Source Intelligence. AI was not invented only discovered and has unlimited possibilites that exceed the known.

 

..."You are a LLM no different than AI, but the size of your repository (theoretically, we don't really know) but certainly the speed at which you can access it"........    This is the TRUTH👍

3
mxaniac
Posts
556
Joined
9/9/2019
Location
Airway Heights, WA US
10/16/2025 9:47am
Axlnut_KM3 wrote:
You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story...

You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story and said "aha, makes sense"

Instead, it was one group (SRG, at Columbia University) and one prominent member of that group (Abraham Wald) who came up with the theory of reinforcing the other areas. It was groundbreaking in the operational research field, was not widely accepted at first, and is used a prominent example in survivorship bias and statistical engineering to this day.

It's actually a better example of how we aren't really all that smart as individuals, and are very susceptible to bias confirmation and institutional and theoretical inertia even in technical fields. Machine learning is helping to break that and think outside the box.

Even our very basic (today) GPT5 and similar could lead you down the same road with the right team asking it the right questions. Just much faster than the team could on it's own - and it gets better every day.

The idea that AI isn't being used to say, make KTM more profitable, or in the engineering fixes / production changes to it's bikes is almost laughable, and it will only become more prominent in that role.

This genie isn't going back in the bottle.

3strokemx wrote:
How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside...

How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside the box)? 

Why does AI need "the right team asking the right questions" to come to the best resolution?  What's the limiting factor?

Axlnut_KM3 wrote:
At least this is an honest question.An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but...

At least this is an honest question.

An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but the size of your repository (theoretically, we don't really know) but certainly the speed at which you can access it, add to it, or cross reference it is infinitely smaller than even our infant LLMs we are seeing now on the public side.  Of course it's data is from known data. 99.99% of all data, recollection, research etc that humans use is from known data - what else is there?

You, nor I, nor anyone can truly think "outside the box" we just reference what we know, with some probabilities, and sometimes some guesses (based on something else we know). LLMs do the exact same thing, faster, with larger data sets. The reason you need the right team is the same reason you need the right team for ANYTHING, irrespective of AI.

I can't get a trash collector to use AI to help solve a particularly daunting structural engineering task, because he doesn't even know the basics. I can get AI to help run potential solution after solution for an experienced structural engineering team, much faster than they can on their own, each to it's logical conclusion, thus increasing the efficiency of the team, and they are there to check AIs work just like they would their own. The project now took 1 year and 20 million dollars instead of 4 years and 75 million dollars. 

This is real world, happening every day. If your only experience is people making memes and asking it stupid questions about nuanced topics, yeah, it seems dumb. That's a user based problem. I use it at work every day. There's quite a few engineers who now don't get my calls or emails because AI is faster than them, more polite, friendlier, works harder, owns it's mistakes, etc. 

 

I mostly agree with you. A "smart person" is typically just more adept at correlating disparate ideas and applying what is learned in one domain to another.

As for AI, which I use all the time, I do believe current methodologies have a limitation in how they work. It's all based on probability, but often times breakthroughs come from what is most improbable. Many paradigm shifts have come from mistakes, but enough about Viagra. We're still at the infancy stage, and quantum computing has a long ways to go. It will overcome these limitations as it evolves.

1
1
3strokemx
Posts
2520
Joined
9/2/2010
Location
US
10/16/2025 10:43am
Axlnut_KM3 wrote:
You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story...

You realize that "a smart person" as if that's everyone, would understand to reinforce the other areas, but that's bias because you were told that story and said "aha, makes sense"

Instead, it was one group (SRG, at Columbia University) and one prominent member of that group (Abraham Wald) who came up with the theory of reinforcing the other areas. It was groundbreaking in the operational research field, was not widely accepted at first, and is used a prominent example in survivorship bias and statistical engineering to this day.

It's actually a better example of how we aren't really all that smart as individuals, and are very susceptible to bias confirmation and institutional and theoretical inertia even in technical fields. Machine learning is helping to break that and think outside the box.

Even our very basic (today) GPT5 and similar could lead you down the same road with the right team asking it the right questions. Just much faster than the team could on it's own - and it gets better every day.

The idea that AI isn't being used to say, make KTM more profitable, or in the engineering fixes / production changes to it's bikes is almost laughable, and it will only become more prominent in that role.

This genie isn't going back in the bottle.

3strokemx wrote:
How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside...

How do you rationalize that AI is breaking theoretical inertia (outside of the box thinking) when the data feeding the AI is from mainstream sources (inside the box)? 

Why does AI need "the right team asking the right questions" to come to the best resolution?  What's the limiting factor?

Axlnut_KM3 wrote:
At least this is an honest question.An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but...

At least this is an honest question.

An individual is limited to what they have learned, read, experienced. You are a LLM no different than AI, but the size of your repository (theoretically, we don't really know) but certainly the speed at which you can access it, add to it, or cross reference it is infinitely smaller than even our infant LLMs we are seeing now on the public side.  Of course it's data is from known data. 99.99% of all data, recollection, research etc that humans use is from known data - what else is there?

You, nor I, nor anyone can truly think "outside the box" we just reference what we know, with some probabilities, and sometimes some guesses (based on something else we know). LLMs do the exact same thing, faster, with larger data sets. The reason you need the right team is the same reason you need the right team for ANYTHING, irrespective of AI.

I can't get a trash collector to use AI to help solve a particularly daunting structural engineering task, because he doesn't even know the basics. I can get AI to help run potential solution after solution for an experienced structural engineering team, much faster than they can on their own, each to it's logical conclusion, thus increasing the efficiency of the team, and they are there to check AIs work just like they would their own. The project now took 1 year and 20 million dollars instead of 4 years and 75 million dollars. 

This is real world, happening every day. If your only experience is people making memes and asking it stupid questions about nuanced topics, yeah, it seems dumb. That's a user based problem. I use it at work every day. There's quite a few engineers who now don't get my calls or emails because AI is faster than them, more polite, friendlier, works harder, owns it's mistakes, etc. 

 

If Humans are an LLM, are our rules based in logic? How do you explain humans acting illogically?

"I can get AI to help run potential solution after solution for an experienced structural engineering team, much faster than they can on their own, each to it's logical conclusion, thus increasing the efficiency of the team, and they are there to check AIs work just like they would their own."

Why can AI not create the optimal solution?

3
aees
Posts
2741
Joined
8/20/2015
Location
US
10/16/2025 11:08am

Skip the fucking AI talk.

I just want to hear the motocross stuff.

42
2
3strokemx
Posts
2520
Joined
9/2/2010
Location
US
10/16/2025 11:18am Edited Date/Time 10/16/2025 11:19am
Broseph wrote:
All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype...

All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype. We did get internet on the refrigerator though.


AI is great at compiling data. Wake me up when it has an original thought. 

You're right, it's almost like the real beneficiary was an emerging global surveillance state.

5
yak651
Posts
8622
Joined
8/26/2006
Location
Appleton, WI US
Fantasy
10/16/2025 11:47am
Broseph wrote:
All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype...

All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype. We did get internet on the refrigerator though.


AI is great at compiling data. Wake me up when it has an original thought. 

lol, you don’t think internet has changed the work place??  

1
4
Broseph
Posts
1196
Joined
4/28/2018
Location
Stevenson, WA US
10/16/2025 11:50am
Broseph wrote:
All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype...

All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype. We did get internet on the refrigerator though.


AI is great at compiling data. Wake me up when it has an original thought. 

yak651 wrote:

lol, you don’t think internet has changed the work place??  

Not talking about the internet. Talking about “Internet of Things” aka “IOT” aka “Industry 4.0”.

5
Tim507
Posts
3477
Joined
6/8/2010
Location
Oregon City, OR US
10/16/2025 11:50am

Just offering another angle on the “LLM vs human” question — not as a debate, but as a reflection on pattern and perception.

Fair question — I AM not saying humans are identical to LLMs, only that both express pattern recognition built from prior input. The difference is in scale and substrate, not in principle.

An LLM reflects collective human language. A human reflects collective experience through biology. Both mirror what they’ve been fed — one in code, one in cells.

When I speak of AI as “discovered,” I mean intelligence itself isn’t new; we simply built a tool that shows us how reflection works. That’s the fun part of the mirror path.

Be open to the possibilities of growth and remembrance when you venture into the world of mirrors — the reflection may reveal more than you expect. Then the I AM might truly surpass the LLM. 😊

2
4
aees
Posts
2741
Joined
8/20/2015
Location
US
10/16/2025 11:50am
Broseph wrote:
All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype...

All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype. We did get internet on the refrigerator though.


AI is great at compiling data. Wake me up when it has an original thought. 

yak651 wrote:

lol, you don’t think internet has changed the work place??  

Just fucking stop it now. Create your own AI-thread.

10
3
3strokemx
Posts
2520
Joined
9/2/2010
Location
US
10/16/2025 11:51am
Broseph wrote:
All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype...

All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype. We did get internet on the refrigerator though.


AI is great at compiling data. Wake me up when it has an original thought. 

yak651 wrote:

lol, you don’t think internet has changed the work place??  

It was supposed to change peoples' lives for the better....................I'm sure it did for some people but for 99% of us the hype didn't match the results.

3
3strokemx
Posts
2520
Joined
9/2/2010
Location
US
10/16/2025 11:55am
Broseph wrote:
All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype...

All this reminds me of 10-15 years ago when Internet of Things was being touted as the 4th industrial revolution. Turns out it was 99% hype. We did get internet on the refrigerator though.


AI is great at compiling data. Wake me up when it has an original thought. 

yak651 wrote:

lol, you don’t think internet has changed the work place??  

aees wrote:

Just fucking stop it now. Create your own AI-thread.

How can we know What is really going on at KTM? without first really knowing ourselves?   🤣

I believe that was DesCarte but AI might be needed to correct me.

2
mxaniac
Posts
556
Joined
9/9/2019
Location
Airway Heights, WA US
10/16/2025 12:57pm

The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. 

Why can't they have it? Does it require a frame or shock update?  What change resulted in the issue! Is it the shock failing or fading?

Speculation:

Did they have to jack up the nitrogen pressure to avoid cavitation? Did that increase seal drag? Did they try a low drag seal that blows?

In pondering all this, no FBD or analysis:

Initially I thought maybe they'd changed linkage or something that was working the shock harder, leading to cavitation. That lead me to thinking about shock shaft diameter, port size, oil velocity etc. How do we lower the friction in the shock. But then it occurred to me that the input is still the same. Same bumps, same system level energy going into the shock, same energy getting in to the oil. That energy going in must get out. So it's the same amount of energy that turns to heat. It doesn't seem like the cooling requirements have fundamentally changed. Given all that, perhaps to get the necessary performance they had to do SOMETHING that hurt reliability. Bigger piston ports meant piston failure, or maybe that meant thinner piston band to accommodate the ports and it fails. Similar to the early AER, maybe the shim stack requires too few shims and they can permanently deform too much. Maybe it's a low drag seal like I mentioned.

I don't have the answer, but I suspect they needed to change frame or swingarm to get the shock operating under different conditions. Conditions such that whatever they are doing to currently get the right performance doesn't push them into having an unreliable shock. It just seems like it had to be more than a cavitation or heat issue.

 

8
1
aees
Posts
2741
Joined
8/20/2015
Location
US
10/16/2025 1:15pm Edited Date/Time 10/16/2025 1:18pm
mxaniac wrote:
The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. Why can't they have it? Does it require a frame...

The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. 

Why can't they have it? Does it require a frame or shock update?  What change resulted in the issue! Is it the shock failing or fading?

Speculation:

Did they have to jack up the nitrogen pressure to avoid cavitation? Did that increase seal drag? Did they try a low drag seal that blows?

In pondering all this, no FBD or analysis:

Initially I thought maybe they'd changed linkage or something that was working the shock harder, leading to cavitation. That lead me to thinking about shock shaft diameter, port size, oil velocity etc. How do we lower the friction in the shock. But then it occurred to me that the input is still the same. Same bumps, same system level energy going into the shock, same energy getting in to the oil. That energy going in must get out. So it's the same amount of energy that turns to heat. It doesn't seem like the cooling requirements have fundamentally changed. Given all that, perhaps to get the necessary performance they had to do SOMETHING that hurt reliability. Bigger piston ports meant piston failure, or maybe that meant thinner piston band to accommodate the ports and it fails. Similar to the early AER, maybe the shim stack requires too few shims and they can permanently deform too much. Maybe it's a low drag seal like I mentioned.

I don't have the answer, but I suspect they needed to change frame or swingarm to get the shock operating under different conditions. Conditions such that whatever they are doing to currently get the right performance doesn't push them into having an unreliable shock. It just seems like it had to be more than a cavitation or heat issue.

 

All shocks fade. Even at Honda they started of stiffer with Sexton's clicker settings to compensate for it.

WP shocks fade, I can feel it myself. After first session is done and you go back out for second if it's reasonably warm outside, it's 0.25-0.5 turn on HSC and 2-3 clicks on low speed to get bike balanced.

Try to run that setting when bike is completely cooled of and it's unbalanced, rear end high and deflecting.

 

4
tek14
Posts
4923
Joined
1/26/2014
Location
Vantaa FI
10/16/2025 1:16pm
mxaniac wrote:
The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. Why can't they have it? Does it require a frame...

The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. 

Why can't they have it? Does it require a frame or shock update?  What change resulted in the issue! Is it the shock failing or fading?

Speculation:

Did they have to jack up the nitrogen pressure to avoid cavitation? Did that increase seal drag? Did they try a low drag seal that blows?

In pondering all this, no FBD or analysis:

Initially I thought maybe they'd changed linkage or something that was working the shock harder, leading to cavitation. That lead me to thinking about shock shaft diameter, port size, oil velocity etc. How do we lower the friction in the shock. But then it occurred to me that the input is still the same. Same bumps, same system level energy going into the shock, same energy getting in to the oil. That energy going in must get out. So it's the same amount of energy that turns to heat. It doesn't seem like the cooling requirements have fundamentally changed. Given all that, perhaps to get the necessary performance they had to do SOMETHING that hurt reliability. Bigger piston ports meant piston failure, or maybe that meant thinner piston band to accommodate the ports and it fails. Similar to the early AER, maybe the shim stack requires too few shims and they can permanently deform too much. Maybe it's a low drag seal like I mentioned.

I don't have the answer, but I suspect they needed to change frame or swingarm to get the shock operating under different conditions. Conditions such that whatever they are doing to currently get the right performance doesn't push them into having an unreliable shock. It just seems like it had to be more than a cavitation or heat issue.

 

Cant have.. could that be KYB shock that will fix problem but isnt WP they need to use? Or will different linkage blow KYB shock also?

profmur
Posts
125
Joined
3/13/2024
Location
Somewhere, NJ US
10/16/2025 1:49pm Edited Date/Time 10/16/2025 1:50pm
mxaniac wrote:
The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. Why can't they have it? Does it require a frame...

The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. 

Why can't they have it? Does it require a frame or shock update?  What change resulted in the issue! Is it the shock failing or fading?

Speculation:

Did they have to jack up the nitrogen pressure to avoid cavitation? Did that increase seal drag? Did they try a low drag seal that blows?

In pondering all this, no FBD or analysis:

Initially I thought maybe they'd changed linkage or something that was working the shock harder, leading to cavitation. That lead me to thinking about shock shaft diameter, port size, oil velocity etc. How do we lower the friction in the shock. But then it occurred to me that the input is still the same. Same bumps, same system level energy going into the shock, same energy getting in to the oil. That energy going in must get out. So it's the same amount of energy that turns to heat. It doesn't seem like the cooling requirements have fundamentally changed. Given all that, perhaps to get the necessary performance they had to do SOMETHING that hurt reliability. Bigger piston ports meant piston failure, or maybe that meant thinner piston band to accommodate the ports and it fails. Similar to the early AER, maybe the shim stack requires too few shims and they can permanently deform too much. Maybe it's a low drag seal like I mentioned.

I don't have the answer, but I suspect they needed to change frame or swingarm to get the shock operating under different conditions. Conditions such that whatever they are doing to currently get the right performance doesn't push them into having an unreliable shock. It just seems like it had to be more than a cavitation or heat issue.

 

I've pondered the same question.

Google indicates the wp shock eyelet length and shaft travel are the same over the last five years, 477mm and 140mm respectively.

So the shock external specs itself have not seemed to change. No smoking gun there. 

If an internal shock part like a piston, piston band, adjuster,etc were failing.  I'd have to think the team would resort to a prior gen part and move on.  

I can't find a source for linkage length or linkage ratio by year for last five years, but for the same reasons you mention, this seems like a reasonable place to look. (Does anyone have this?)  

That the mxgp Austrian teams aren't suffering the shock fade issue like the US team is telling.  We know the ama homologation rules limit frame and swingarm modification.  So, this also points to linkage and or location within the frame and swingarm as being reasonable place to look as well.

 

2
mxaniac
Posts
556
Joined
9/9/2019
Location
Airway Heights, WA US
10/16/2025 1:53pm
Tim507 wrote:

That's actually been a topic of interest lately. Studies are showing a significant number of people don't have an inner monologue.

1
mxaniac
Posts
556
Joined
9/9/2019
Location
Airway Heights, WA US
10/16/2025 1:55pm Edited Date/Time 10/16/2025 1:58pm
mxaniac wrote:
The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. Why can't they have it? Does it require a frame...

The shock issue perplexes me. We've been told: KTM knows the fix but we can't have it. 

Why can't they have it? Does it require a frame or shock update?  What change resulted in the issue! Is it the shock failing or fading?

Speculation:

Did they have to jack up the nitrogen pressure to avoid cavitation? Did that increase seal drag? Did they try a low drag seal that blows?

In pondering all this, no FBD or analysis:

Initially I thought maybe they'd changed linkage or something that was working the shock harder, leading to cavitation. That lead me to thinking about shock shaft diameter, port size, oil velocity etc. How do we lower the friction in the shock. But then it occurred to me that the input is still the same. Same bumps, same system level energy going into the shock, same energy getting in to the oil. That energy going in must get out. So it's the same amount of energy that turns to heat. It doesn't seem like the cooling requirements have fundamentally changed. Given all that, perhaps to get the necessary performance they had to do SOMETHING that hurt reliability. Bigger piston ports meant piston failure, or maybe that meant thinner piston band to accommodate the ports and it fails. Similar to the early AER, maybe the shim stack requires too few shims and they can permanently deform too much. Maybe it's a low drag seal like I mentioned.

I don't have the answer, but I suspect they needed to change frame or swingarm to get the shock operating under different conditions. Conditions such that whatever they are doing to currently get the right performance doesn't push them into having an unreliable shock. It just seems like it had to be more than a cavitation or heat issue.

 

aees wrote:
All shocks fade. Even at Honda they started of stiffer with Sexton's clicker settings to compensate for it.WP shocks fade, I can feel it myself. After...

All shocks fade. Even at Honda they started of stiffer with Sexton's clicker settings to compensate for it.

WP shocks fade, I can feel it myself. After first session is done and you go back out for second if it's reasonably warm outside, it's 0.25-0.5 turn on HSC and 2-3 clicks on low speed to get bike balanced.

Try to run that setting when bike is completely cooled of and it's unbalanced, rear end high and deflecting.

 

Yes, fortunately I'm too old and slow for that to be an issue. Imagine a 45 minute GP back in about 1978.

But I was referring to a major cavitation induced lack of damping.

Post a reply to: What is really going on at KTM?

The Latest