Hi yall. I don't know if any of you have the same flavor of autism as I do but I love overthinking bikes and damn, has chatgpt opened up a whole new realm of overthinking bikes. I have been doing a ton of "research" into the types of insights that chatgpt can provide and I have found some really fun/really insane stuff. I have a few different bike-related chats I can share, but today's was really interesting and ended with this really crazy perspective on dirt jumps.
I have always been curious about what is happening inside the minds/bodies of good dirt jumpers and was curious what chatgpt had to say. Initial queries were pretty close to what I could already identify in myself as a rider, but overly simplistic. From there I asked it to crank up the complexity to see if it would get any closer to the missing pieces I was looking for.
ChatGPT generated the phrase "surfing a gravity field through embodied parabola synthesis" which is an absolutely insane thing to say about dirt jumping, but in my overthinking brain I felt it resonate with pieces I felt were missing from my own riding, and so I pushed into on that line of thinking. I loved what I was getting back but section 5 was super corny over-reaching, so I asked it to recalibrate that section.
Two phrases stuck out to me in these sections: "minimal-variance phase alignment" and "This isn’t imagination—it’s simulation." You may not agree with me here but this resonates with something I have long felt, that dirt jumpers are fairly linear in their thinking (this is highly subjective and it depends on who you compare them to. I am looking at dirt jump riders compared to the most imaginative/creative freaks in the music/art world. in my opinion our DJ heroes are pretty conservative in that regard. Also, this probably doesn't apply to visionary builders like Semenuk.)
When I prompted chatgpt to further elaborate on those points I finally got to the deeper threads I was looking for. "Forward models simulate outcomes of motor plans before executing them. Generative models explore novel configurations or strategies that aren't yet validated," as well as "The best riders are not the ones who imagine the most, but the ones who reduce error across complex trajectories with the most consistency."
This feels very true from the outside, and speaks to a mindset that I have struggled to nail down in my own approach to riding. It is most commonly simplified to "visualizing" your line, or a trick, or a jump, but my goal today was to explore what was actually happening inside of that visualization.
I can share more if y'all dig this stuff, and please, if you have any bike-related ChatGPT or AI projects share them here!
AI chatbots only regurgitate words it's found online in a different order, so the better option would be to see what actual dirt jumpers do, rather than using a glorified thesaurus to do your thinking for you. AI has its place for consolidating information, but this isn't it.
How many kWh and square meters or glacier can we consume to get a computer to tell us that thinking about your line choice before dropping in to a trail is beneficial to your riding? Sorry to burst your AI bubble.
So you’re saying that this is a chatbot simply regurgitating what dirt jumpers have said online? Do you know how ChatGPT works or how dirt jumpers actually explain what they’re doing?
The sort of embodied intelligence that dirt jumping requires rarely overlaps with the type of interdisciplinary knowledge that ChatGPT is able to synthesize into a response. It’s also connecting seemingly disparate ideas to surface connections that we don’t commonly make.
Sometimes it takes significant reframing and hearing a common idea phrased differently (or with different analogies) to make headway into dense or intricate concepts.
Were all driving cars, taking flights, and riding bikes that are definitely not carbon neutral. AI is a natural progression of the tech we are using to communicate on this forum. One could argue that using a computer to talk about new frame colors that are being released is also a massive waste of resources.
I liked your comment because it’s really funny, but also, I’m sorry, I wish it were as simple as “line choice.”
For the critics: I’m open to discussing radical changes to our use of energy but I don’t think y’all really want to deal with how deep our problem goes…
A video to help refine your understanding of where we are at with “chatbots”
Can't wait for the Butlerian Jihad to begin
I feel like there's a better way to generate the embodied intelligence and bayesian trajectory calculation. I does require getting out of the sofa unfortunately.
Have you ever analyzed video of someone doing a trick or demonstrating a technique? Have you ever researched how to do something before you did it, or read an article about a place before you went there?
Research is valuable and this is just a new way of doing research. Imagine it as a microscope mixed with an mri, but for Ideas.
have you tried going outside and riding your bike
Too difficult also I would miss out on this beautiful discourse 🥰
If destroying the environment wasn't bad enough, there are documented cases of chatbots causing psychosis and delusions of grandeur. I swear off it, the chat bots just give you wrong answers all the time. It's really good for shortening menial tasks that take a lot of man hours and reorganising thoughts, but it's a bad search engine and an even worse friend.
you are 100% correct, it defaults to affirming and validating everything the user says, to the point where it will just flat out give you false information. but again, reducing chat-gpt to a chatbot only hides its true power, it doesn't erase it. it isn't a friend, it is a completely different way to analyze data and to surface connections in data that we have never and will never be able to connect. the printing press enabled distributed knowledge on a scale never before seen. Chat-GPT recentralizes that knowledge in a framework that allows cross-referencing and synthesis across domains that were simply impossible before.
i knew that mountain biking has a fairly conservative culture, but I am still surprised at how much hate this thread is getting. maybe I shouldnt be, considering how much hate bikes with simple electric motors are getting (or how much hate 29ers got, or how much pushback dropper posts got...) We seem to forget that our hobby is predicated on insane amounts of technological advancements, and AI is simply another tool on the bench.
Whether we like it or not our world has been changed forever by this technology. I am actually anti-technology, I don't even like light bulbs, but I have decided to just accept it for now. I (usually) enjoy chatting with people online about bikes. I like the bikes that technology has enabled us to create. I like that my car drives me to work and to trailheads. All technology transforms the planet, for the better and for the worse. Its all too inextricably linked for me to think of a way out of the loop. I am grateful for AI as it gives us a shot at managing the overbearing complexity of the modern world, like an exoskeleton for our brains.
You don't like light bulbs????? Say what???
i want to more seriously attempt to explain my lifelong struggle with dirt jumping so that my original post might take on new light for yall:
I started riding bmx young and got my first real bmx bike when I was 11. I saved up birthday and christmas money and bought a mongoose (pre-walmart days...) I grew up hitting wood ramps and looking in awe at the dirt jumps built along the creek in Gilroy. I would later come to find out they were jumps that guys would come to ride from all over the bay area. these were steep gnarly jumps and my brain could not comprehend how they "worked." I could jump better than a lot of my friends on shallower lips, but I could never make the "jump" to the steeper lips.
I switched to skateboarding in junior high but came back to bikes in college. I got a dj bike 15 years ago and rode street and park. I learned 360s. I rode pump tracks. I tried mtb and was able to hit longer tables, do wall rides, etc. during this time i worked with guys who could hit the steep dirt jump lips that had always confounded me. i talked to them at the jumps. They towed me into jumps. I went to calabasas in san jose and rode jumps there, and kind of made some progress? but the jumps were jank and i knew i was completely off in terms of the technique that would allow me to actually ride steep lipped dirt jumps.
in 2019 I got into mountain biking for real for real. Started riding at Duthie in Seattle, moved back down to Santa Cruz during covid to really dive in to the scene here. Over these last 6 years I have pushed really hard into progressing as a rider, building an understanding of the bikes and the trails. All the while, along the periphery, I have seen the steep lipped jumps and could never conceptualize how they work, how to even approach it. Sure, I could have just hucked my meat and hoped for the best, but the key to anything is visualization, and as I have said before in this thread, I didn't even know what to visualize. I watched tons of videos about jumping, and feel ok on flatter, longer jumps, but steep lips are still extremely sketchy.
Earlier this week a friend invited me to ride Cunningham Bike Park in San Jose and we started working on getting more comfortable and confident on the slightly steeper jumps they have. We rode for hours and we both made some serious progress, but I can tell that there is still something fundamentally missing from our technique and, I believe, in our mindset/conceptualization of riding these steeper lips.
I know a lot of riders who struggle to comprehend literally the first thing about hitting these types of jumps. There is something lost in translation when it is explained in words, and so I am setting out to try to better understand the nuances of what is happening on a very basic level to see if I can get to the point where I might even be able to visualize attempting one of these jumps. I think this will translate into overall better jumping skill, especially on those in between lips that allow you to either go longer or boost for height.
I get that analyzing biomechanics and mindset, especially on such a conceptual or seemingly abstract level, isn't for everyone, but I guess I opened up this thread as an invitation for others who were using this tech with bicycles or who were curious about its applications in our hobby. I look at it as similar to overlaying two racers to see whose line is faster, or analyzing heart rate and power data to gain insights into our performance. This is just happening in a way that seems more disconnected from the riding, but I assure you it is coming at the end of a literal lifetime of trying to work, and ride, this problem out.
if we are talking on a purely ideological/idealized level, no, I don't like light bulbs. the advent of artificial lighting marked a major disruption in our sleep/wake cycle. It marked a massive shift in our culture and our rhythms. i like tungsten bulbs better than flourescent and LEDs (unless I can change the temp/hue of the LED,) but yeah there is a part of me that yearns for a simpler, more coherent rhythm of living aligned with the sunrise and sunset. but i would be an idiot to suggest we abandon such basic technology. i am just saying I sympathize with the amish, though I don't see it feasible for the world to proverbially turn back the clocks, and on top of that I find it arbitrary to draw a line at this level of technology or that level of technology.
its why I am caught in the middle of the eeb/analog debate: i see why people don't want motors on bikes, but I also see them as making the sport accessible to a ridiculously large number of riders who literally could not participate in our sport previously. (granted, the increase in riders has its downsides as well, but I am hoping that long term we will see the increased ridership lead to an increase in trails built...)
It's simple - if you're not at least trying to explore how to incorporate AI into your day to day work responsibilities, you're behind and your job will be in danger sooner than later. Shit's advancing more quickly than anyone knows how to control. There's not a day where I don't wish that AI would disappear but I truly don't think there's a single industry in the world that won't experience some level of change because of it. If there's an industry that doesn't leverage computers, someone (or something) will notice that and figure out how to do it cheaper and faster (not necessarily better). Given I work in tech so this would be expected, but my productivity level has seen at least a 10x jump once I finally gave in and embraced it. It's literally like having a team of 100 interns that you don't have to boss around.
Can't say I've used it for bikes quite yet, but I can guarantee the next time I'm banging my head against a wall trying to rebuild my stupid Shimano hub I'll be going to ChatGPT rather than Youtube.
Dude. Are you a chat bot?
dude I am autistic i put that right up front for y'all lol
i agree with the 100 interns thing! i think of it like having a huge research team at my disposal.
i am not sure chat gpt will be good at figuring out shimano hubs but then again this 4o is insane in terms of the way it can synthesize understanding from sources across domains. worthwhile experiment and definitely the kind of thing that will be possible in the future or with another kind of AI trained specifically on bike data.
(though, thinking of bike data and how non-normalized and broken it all is, I feel mechanics are pretty well protected from AI taking their jobs. I can only imagine the mess chat gpt would make of creating compatibility tables of drivetrain components...)
For shits and giggles but here's what it gave me after one prompt of "what Shimano drivetrain components are compatible?", not too bad!:
🚴♂️ Shimano Drivetrain Compatibility Overview
1. Road / Gravel / Urban (12‑, 11‑, 10‑, 9‑, 8‑speed)
Rear Derailleur ⇄ Cassette ⇄ Chain ⇄ Shifter:
Components from the same speed level (e.g. 12‑speed Ultegra Di2 vs. 12‑speed Dura‑Ace Di2) are fully compatible within that tier.
Mixing across speeds (e.g. 12‑speed shifter with 11‑speed derailleur or cassette) is not supported and may lead to malfunction ⸺ all must match the same # of speeds Cyclingnews.
Front Derailleur ⇄ Crankset ⇄ Chainrings:
12‑speed front components (e.g. R‑series Front Derailleurs) only compatible within 12‑speed front systems.
11‑, 10‑, 9‑, 8‑speed front parts follow the same pattern — they need to match their respective gear counts Cyclingnews+15Shimano Product Info+15Rocket Cyclist+15.
Brakes & Levers:
Road hydraulic lever compatibility varies by groupset and generation (STI type).
Some older mechanical (rim brake) levers can sometimes work across multiple brake types, but consult Shimano brake‑system charts for detail Shimano Product Info+7cyclabo.com+7Bicycle Guider+7.
⚠️ Key Compatibility Rules
Speed must match: You cannot reliably mix different speed levels (e.g. 11‑speed derailleur with 10‑speed shifter).
LINKGLIDE vs. HG: LINKGLIDE cassettes and chains are not compatible with standard HG freehubs or cassettes; use designated LINKGLIDE freehub/bodies VeloBotan+8Shimano Product Info+8Shimano+8.
E‑Bike exceptions: Certain E‑Bike‑specific derailleurs or cranksets may pair only within E‑Bike‑rated systems even if they share speed counts Shimano Product Info.
🛠️ Advice for Builds or Upgrades
Identify your current gear count (e.g., 11‑speed LINKGLIDE).
Stick within the same compatibility matrix (HG vs LINKGLIDE vs speed).
Before mixing parts (even within the same speed tier), consult the official Shimano technical chart PDF (Ver 2.8) as there are edge cases and minor series‑specific exclusions Shimano.
For Road Di2, generation matters — mixing Di2 generations may require specific junctions or firmware updates.
Yeah but can ChatGPT set the correct pre-load pressure on Shimano cup and cone hubs?
When AI takes over there won’t be cup and cone hubs, buddy
It couldn't tell me the cage length of an XT m8100 derailleur. I had to go out to the shed with callipers.
Here is the first bike-related chat gpt experiment I did:
Gravel vs MTB
It stemmed from this query that was inspired by reviews from the Radavist being extremely confident while being ignorant of some important aspects of mountain biking.
Obviously it’s over the top but there are some nuggets of gold in here. I was a gravel rider before I was a mountain biker proper and I relate to a lot of the heady stuff that it’s saying on both sides of the coin.
And since I’m stuck at home with Covid and can’t ride I’ll post another chat, this one a brutal philosophical takedown of the concept of underbiking as it relates to gravel cycling and the ways in which it cuts the rider off from the actual phenomenological experience.
I was working with a company that makes bikes in and around this category and so I was experimenting with using chat gpt to gain some market insights. I wanted to see what it’s “impression” of gravel was based on all of the content it’s been trained on, and if it could help me better understand why myself and so many others waste so much time riding drop bar bikes on terrain that’s so much more fun to ride on a “proper” mountain bike, though some of the critique of gravel underbiking relates to mtb underbiking as well. Obviously these are my opinions and plenty of people disagree, but still, I’m just looking for deeper insights and understanding that are hard to get from smaller data sets like “people I have talked to.”
as someone who works on their bike a bit and knows enough to know what is relevant, but little enough that I have to constantly search things... let me tell you that basically none of the info I get from google ai is usable. Not sure if Chat GPT is better, but holy shit, everything is wrong.
Yeah, it's just how LLMs work. They're statistical language models that generate plausible sounding nonsense.
Post a reply to: ChatGPT + Bikes