Ha! This is pretty well known on the science end of things, though I always like it when other people learn this. It’s why acknowledging uncertainty and not having strong opinions on very specific training matters is probably the best indicator that someone knows what they’re talking about.
You can likely draw some inferences from large datasets like TR has. However, data drawn from databases or registries is fundamentally low-quality evidence that is pseudo-experimental at best.
You can definitely learn from it, and that would be easier with ML, but it’s just lots of data. Not high quality experimental data. It’s the same way how registry data does not supplant the need to perform clinical trials in medicine. The only way we will have certain conclusions is by conducting high-quality primary research. Not retrospective data mining existing data.
There will be fun insights to be had. But I don’t expect it to change the face of training. The answer is still going to be to ride your bike more.
many top-level coaches are apparently completely unperturbed by facts and new insights, and instead want to stick with what they have learnt 20 years ago.
Many athletes as well. And people in general. It’s another reason why the fundamentals of training are unlikely to change significantly. Most people just do whatever they feel like doing and then find creative ways to justify it.
Maybe its really about how fast you pedal your bike. Is high cadence climbing gonna make a return??? That was of course what helped Lance win 7 tours…or at least that is what all the forum chatter was back in the mid 00’s
No doubt there’s some truth here but it is also a pretty arrogant take on the researchers part. Coaches are on the front lines, they know what works because they have seen the long term results. How many researchers collect data for an athletes entire career? The reality is coaches know what work, physiologists try to figure out why. Science has never driven training… it’s the coaches and athletes.
Here is a great article on the history of long distance training by Steve Magness. Reality is there isn’t really anything new… but is rather more of a cycle. Maybe we are just at the end of that 20 years and about to see something old become new again.
Yes, but I was still surprised by what is actually not yet established. The discussion centered around super compensation, and according to this researcher, super compensation has been established in terms of glycogen processing capabilities (I cannot recall the specifics with sufficient certainty).
I have colleagues in astronomy, dealing with large datasets or datasets with large margins of error is by no means new. Although I have to admit, it is a very different way of thinking.
In sports science the N is typically very, very small, especially once you go in-depth and e. g. insist that people use that fancy SRM lab-grade ergometer. The sheer number of questions you could answer with TR’s database is quite intriguing.
I noticed this in a few places (as a layperson). E. g. Alex Dowsett complained that some of his teams were clueless or didn’t care about about basics like aerodynamics and nutrition. And many of the athletes who are pushing the boundaries have taken an active interest in optimization.
Thing thing is, they all know what works. And it’s not super complicated. I’d refer anyone to the last Empirical Cycling podcast where Kolie and Rory sketch out two basic build plans: endurance (base), threshold - build out TTE, VO2, race prep (short intervals), taper. It’s classic build the house, raise the roof (vo2) stuff. Kolie even says that endurance and threshold (of any variety (tempo, sweet spot, ftp)) is the ‘peanut butter and jelly’ of training.
Everything is made out to be so complicated but if the average amateur rider just did the PB&J of training they’d reach 95% of their potential without bespoke coaching, magic intervals, crazy complicated over/under sets, AI, HRV, whoop, or any other fancy gizmo.
Magness has finally launched a YT channel and he’s been doing these 30 minute deep dives on training. They are really great IMO.
Recently, I really liked his mini lecture on the history of endurance training.
I think it applies to cycling as the basics are pretty similar.
I also want to say that I suspect that the value of training data is probably pretty low. TR, TP, Zwift, Strava, Wahoo, Garmin, Intervals.icu, etc. all have massive amounts of training data.
How is anyone monetizing it?
Is the value of the data of 1M athletes any better than that of 500 athletes?
These companies have had our data for years, if not decades and they have come up with zilch to actually improve training. TR has had to come up with AI FTP guesstimating because people don’t want to test. They had to come up with red light / green light because people can’t tell that their legs are fried and that they should hold back.
So, I’m not optimistic that any company will come up with some game changing software or device that fundamentally changes training.
Hard agree. Apart from the fundamental principles that “more volume is better than less volume”, “at some point you’re going to.have to go harder”, and “you can only do what you can recover from” pretty much (or absolutely) nothing in big data will help you if you’ve got to the intermediate stage and are looking to push on from there.
Humans are weird. They respond very differently to training stimuli. What’s perfect for Rider A might not work at all for Rider B, even if they’re practically indistinguishable. Ultimately, we’re all an experiment with 1 subject and we’re all going to have to FAFO once the noob gains have been milked.
I disagree with this, because it’s based partly on this premise:
I don’t hate AI chatbots or AI tech support because I want a human relationship or need a real person interaction. I hate it because it sucks. I hate it because it can’t understand what I’m asking or because it’s not actually AI it’s just using basic algorithms. I hate real person tech support just as much when they have no clue what they’re doing. It has nothing to do with AI vs human. It’s about how useful it is right now.
I think AI has a long way to go. Which is why I disagree that we’re anywhere remotely close to the final form of AI anything. I’m sure some people will still want that human interaction but I do think that AI coaching could be incredible in the future. Just think about 24/7 coaching vs a real person. Would you wake up your coach at 3AM to ask questions? Do you even have that ability (their personal cell)? What about instant nutrition answers. Like, hey AI Coach, I have my planned workout today and here’s my fridge, what should I eat? And it knows what workout you have today and tomorrow, what the nutritional needs are for that workout, what the athlete has eaten so far that day and the day previous (let’s pretend the AI coach app has a food diary function), so it can use all the available info to give you much more targeted food goals. I really feel like the true final form of AI coaching will be like having a top level coach as a 24/7 always available sidekick.
I think somebody mentioned it, but I think the future of training is going to be more wholistic. Cycling, strength training, nutrition, sleep, stress, psychological.
I think the nutrition part could be pretty cool for fueling workouts and races. Like have a full food diary, so it knows what you’re eating and when you’re eating. And it has a planned time of the workout, so it can give you recommendations or advice on what to eat when. So it’ll pop up in the morning and say: your workout is Threshold at 5PM, here’s a meal plan for the day. And it will adjust it based on what you eat so if you have to skip breakfast because you’re running late for work it will give you a reminder and suggestions. It can also use your motion data throughout the day to adjust your macro and calorie recs. And if you didn’t eat enough the day before it can adjust.
And then use workout data and nutrition data to find holes, like this day you failed your workout but looking over your food diary you didn’t have any carbs. So it uses your diet and training plan together to improve. So the idea would be that it learns how different foods affect your workouts. or learns when the best time to fuel your workouts is. Also uses the current workout to give fueling suggestions during the workout and after.
A big one for me is weight loss. I’d like to lose some weight but I don’t want my training to suffer. So it would use different phases of the training plan to change caloric intake so that you’re not negatively impacting your key workouts.
I’ve had a coach before that I could call or text at any time. I didn’t, because I don’t really need their opinion 24/7.
Training isn’t ever an emergency.
You may personally like the ability to ask someone a question at 3am, but I can’t for the life of me think of a situation where not getting a response immediately will have any significant impact on your training.
Same with nutrition. If you’re just talking from a performance perspective, like 99% of nutrition is simply “eat enough.”
So you may like having a product that does this, but it won’t have any noticeable impact on your training, which is more what I was trying to refer to in my comment.
Also, LLMs already exist which can basically do what you’re suggesting - answering a question in the middle of the night, asking what meals you can cook based on available ingredients. Just play around with ChatGPT 4.
Though I also wonder whether simple questions like you suggested are more effectively answered by googling and learning about the topic, than from asking an LLM.
These people do exist, not many but a few. They would be schooled in all those aspects. The ambition was to be a coach to top level pros. For many reasons it never happened. You can find them at top level fitness centers. They can be pricy but then we spend big money on marginal gains. 1st world problems.
I agree with the idea of this, though I don’t think it’s 99%. Maybe 70-75%. Which is why I think the nutrition aspect of an AI coach could be useful if it improved. Like I can wake up and eat a 1200cal Big Mac for breakfast and then nothing all day. Or space out the calories, eat some protein, and favor simple carbs in the afternoon. Which is going to be better for my 6PM workout? I doubt the difference is only 1% like you say. So it’s not just “eat enough” and you’re good.
It’s not that it’s a training emergency or that it needs to be answered right now, it’s that it could. It’s like having my coach sitting next to me 24/7. And they never get annoyed or tired. So if a thought or question pops up in my head, I ask it and get an answer. Unless you’re Tadej, I’ve never heard of any coaching service ever that could offer that. No coach would unless they’re getting paid crazy amounts.
My point was actually that human interaction isn’t as important or desirable as you’re making it out to be from your comment. Like, it’s not the human interaction that matters to me. I’d choose an AI coach over a human literally any day if the AI had better capability. I don’t care about the human aspect, I just want the best training.
They’re pretty limited. Which was my point. You can only get so far with them and they don’t always give the best answers.
Again, I think you’re missing my point, this isn’t Googling training philosophies. This is about training future. The topic is what you hope to see from training services. And I think that AI will get better and be way more beneficial than a human coach. It’s just an opinion.
Edit: I had some more thoughts based on what you said with “simply eat enough.” Which I think highlights part of my point and the usefulness of a holistic app. What is enough? How many calories is that? What about carbs? How many carbs is enough? Putting apart the macros and timing bits, a simple question like how much food is enough isn’t always so simple. Am I trying to lose weight or gain? What phase of training am I in? Do I need extra calories or can I get by with cutting some? Now I could research this and look into every workout I have planned for the week and calculate estimated calorie burn based on my wattages, but that’s kind of the point of the app. It would look at the workout and my power and give tailored suggestions. There are already apps like Eat My Ride that do this so the next logical step is to put those features together with a training plan as one app.
Again, all of this is a wish list of sorts for future training. Mostly because I have struggled with eating in the past. Gaining and losing weight. Not fueling correctly. Nutrition is probably my weakest point in my training. So just “eating enough” never really worked for me. I need a more structured plan of attack.
