I couldn’t agree more. Part of the reason we’re even having these discussions around various commentators and perceived problems with the current TR platform is because an undisclosed number of people weren’t happy. For whatever reason, the current platform didn’t work for them. Numerous issues have been referenced, time and time again.
I’m willing to bet that ML/AI will not work for everyone. There will be a raft of reasons why. Of those people, I highly doubt that everyone will take an objective viewpoint. It’ll just be like an echo chamber.
I’m not discounting anybody’s opinion or experience, I’m simply trying to keep an open mind to range of possible outcomes
I watched the video and I was definitely put off by revealing a personal convo/interaction with Nate that sounded private.
I’m no TR fanboy and have only used it sporadically over the years - mainly in the winter for motivation to be fit going into the outdoor season. When I watched the TR YouTube video, my takeaway wasn’t that AT was going to replace coaching. My interpretation was more that it could replace a good segment of the coaching community that really only use cookie cutter plans anyway. When I say ‘could’ I assumed Nate was saying the potential was there, but won’t be right away.
The other issue the guy in the video doesn’t seem to understand, is there is such a tiny portion of athletes out there that are even willing to pay for a coach at all! I firmly believe a good coach is better than AI for the reasons he mentioned, but I first have to find a good coach that I gel with, and I have to be willing to pay for that! A good coach will not be cheap and there is no way I would put that much money toward a hobby.
It makes me think about how computers changed chess. Thirty or 40 years ago, everyone had ideas about what the “right” moves were. Computers came along and exposed some weaknesses in the old ways and showed that a lot of those old ideas were basically bro science. Now the best humans can’t beat the best computers. But the computers didn’t completely replaced the coaches and teams that support the top players. It elevated the overall level of competition because people are working with what the computers know. Garry Kasparov wrote a book about all this; it’s good.
Anyway, shifting over to bikes, I think this video (like some other videos) misuderstands what TR is. It is not a full replacement for coaching. I think AT is a step towards being a coaching replacement for a lot of people. But like Nate says, and this guy says a lot in this video, there’s still a bunch of stuff that coaches do that computers can’t. I think in the future coaches will be flipping levers to have the AI pick the workouts/plans that fits their training and racing strategy. But they probably won’t be designing workouts and plans like they have been doing.
In the end, I think these sorts of reaction videos that seem to deliberately miss the point are kinda dumb. I guess I’ve watched a few, though, so maybe that says something about me.
That my my take as well. I see TR moving in to compete with coaches who are low-touch (i.e. infrequent emails, texts, plan adjustments, discussions, feedback, mentoring, etc.) and selling cookie cutter plans. While I think they are in the same general space, I don’t really view TR as a competitor to high-touch coaching companies / services like EVOQ. High-touch and personal attention == $$ and my expectations for the contact and services I receive are going to scale with the amount of money I am paying. My expectations are pretty low for $20/month. For $600/month my expectations in terms of plan development / customization / adjustments, email interactions, texting, video chatting, etc. are much, much higher.
I’m choosing to not watch the video. I’m over all this debating. I think this speaks to the gaining popularity of TR. It’s obviously caught the attention of those trying to make a living at this. I’ve been super impressed with all the work and passion Nate and the team have put into the product. It has grown leaps and bounds all starting with the calendar. It is impossible to make everyone happy but it seems they sure to try at TR.
One of the impressions from reading a lot of threads on this forum is that a significant number of people do view original TR as the coach. A lot of those threads are about ‘failing’ workouts, burning out, inaccurate FTP estimation, or having other problems as they try to complete the plans 100% as proscribed.
The other side of that coin is I see people who view TR as what I would call a ‘self-coaching tool’ that does a lot of the heavy lifting around developing and updating plans, but these individuals customize the plans to their needs. I don’t recall seeing many threads from these kinds of customer complaining about the same types of things as the ‘TR == coach’ crowd.
@bobw Of coarse. I’m not disagreeing. However, to be fair, I did write it’s not there yet. Meaning I don’t think it can do as good of a job as a good coach. Maybe some day. Will the current iteration be acceptable for many? Certainly. But, it’s not a coach nor do I think it was intended as such. A starting point as you wrote.
I don’t care if it ever replaces coaching or not. The question that each rider will have to figure out is does it do better than what they were doing before? Is it doing a better job than a coach? If I had control of my schedule and was really interested in performing to my highest potential in 2021 I hire a coach. 2031? 2041?
I’ll never hire a coach because I don’t race anymore and I don’t need motivation, training is very intrinsic to me. I hold myself accountable and I adjust my plans as needed. So for me personally AT is a great coaching proxy, especially for the price of TR. The entire podcast puts Nate’s comments into better context, and its irresponsible to base an entire opinion of what TR is trying to do with AT and its relationship to coaching on just a snippet.
Very interesting that Nate revealed Adaptive AI, a huge company secret, in order to not make Brendan feel like information was one way.
In the first few minutes Brendan, revealing a private conversation, laments that Nate didn’t say in public what he had said in private: ‘coaches are good for skills, accountability and mental coach’.
But literally from the podcast:
And, um, uh, that’s the longterm goal on this absolutely.
Is to beat a human coach for picking workouts.
And I believe that what we’re, uh, announcing today beats a lot of them.
I don’t have the data to, that’s going to beat everyone, but in all situations, um, But we’re going to try our best to get there. Uh, and I think for, I think it beats a lot of coaches, sorry, coaches,
I still love you for skills and accountability and all this sort of stuff.
I didn’t really feel like continuing watching the vid after this misrepresentation.
Anyhow, great publicity for TR. Brendan has to first inform people that AT has arrived and then has to explain what it is. Presumably he goes on to contest the claims, probably leaving viewers with the undeniable feeling that he sees TR as a threat?
I like Brendan and I think that he has made legit and respectful criticisms of TR in the past but this video was bad. His interpretation of Nate’s “energy systems” blunder is so uncharitable and he also misrepresents what Nate said in the podcast. He claims that Nate said that coaches are good for skills etc in a private conversation but not on the podcast, but actually he did say it on the podcast. He was also critical of ML which is fair, but Nate also said that this entire thing is a long term goal (big hairy goal or whatever) which is supposed to sound a little crazy. I believe coaches can still relax for a bit. ML won’t replace them tomorrow, but in 20 years? Maybe.
AI/ML is going to shake up MANY white collar professions in the coming decades.
To think that cycling coaching will be spared is just comical.
If anything, a high level coach could utilize AI/ML to expand his client base tenfold without having to hire people. The AI would do all the grunt work while the coach would make tweaks and just make sure it’s not doing anything stupid.
I’m a software developer, and I’ve worked a tiny bit in the ML space. Talk to a dev who works in this sphere and I think they will be significantly more pessimistic about timelines than people here. AI is here, but it hasn’t arrived yet. Check out stuff from openai general intelligence model. It’s amazing but there are often extremely unintuitive mistakes made bc the model has self-trained in a way that humans find utterly befuddling.
I mean look at Tesla and FSD tech. While I find it amazing and fascinating, it doesn’t take much to figure out that it behaves in extremely unhuman ways in some situations. I’m also super excited about adaptive training, but I think it’s a little naive to think that everyone who uses it will be blowing others’ legs off in a year. Anyway, I think genetics will remain the largest indicator of performance. Pretty sure Egan Bernal isn’t worried about AT
These youtube replies are in line with the original statement of “in 20 years, ML will replace coaches for [Z]”. Both sides have skin in the game, and push their own agenda. I can personally attest to hearing “in [multiple of 10] years ML/AI will make [Y] obsolete” 20+ years ago - typically it was coming from people who didn’t know what ML was in the first place, but not always. Algo-driven coaching is not new either - take a look at xert as an example.
If you are trying to use ML as a solution for a problem at hand, 90+% of your success depends on data you have (both quality and quantity) - this is one of the reasons why Goggle, FB, and everyone else is trying so hard to get as much of it as possible. And here is the kicker - by the looks of it, TR’s data is very skewed and hence not very good. And TR, instead of fixing it (by providing ways to generate more data in areas they are blind in, which they could easily do in a week by publishing a few more highly requested plans), decided to make a stand and came up with metrics that proved there was no problem in the first place. I can relate and empathize, and hope my experience is not applicable here, but damn my teams and myself suffered in a long run when I did that.
On the other hand, what the hell do I know…I am just reading tea leaves.
It seems like the only reason that TR has had to implement adaptive training is because people can’t control themselves or follow directions. If you listen to the podcast, it’s always ‘back pedal’, ‘lower intensity’, ‘listen to your body’, etc. But you continue to have a certain A type that will keep pushing and keep digging the hole until their legs fall off. Or they follow a plan with 3 or 4 days of intensity but they don’t swap out workouts when they ride outdoors so they end up with 5 or 6 days of intensity until their legs fall off.
The other issue is not being able to properly asses a rider’s FTP. They go to shorter tests because people don’t want to do a longer test but you get half the riders outside of the bell curve. So for an over tester, adaptive training will be constantly lowering intensity because of rider feedback? Will Adaptive Training ever suggest, like a coach, ‘hey pal, I think we need to a proper long MLSS test’?
How will Adaptive Training deal with people that are loath to actually lower their FTP? We’ve seen the chart where a lot of people have a 300 watt FTP and much fewer have a 290. People want the stupid number for bragging rights. How will adaptive training deal with this? It seems like they should go mFTP but then the system will need some max efforts to model it.
Is TR’s adaptive training going to implement a PDC and analyze it? Are they going to do something like iLevels where interval intensity is customized to the rider?
I can’t get over guys like Tim Cusick or Jim Miller that have said that pretty much all intervals accomplish the same thing. Maybe time in zone is the most important thing? It makes you ask the question - what is the AI going to figure out? Will the AI actually map out different interval progressions based on the athlete’s profile?
Sometimes it is worth taking a step back and looking at the big picture. TR, Sufferfest, Zwift, and perhaps even Peloton, have done a great job “growing the pie”. I am certain there are many recreational cyclists who have signed up for these apps and grown tremendously as riders because of them. For many, that is likely enough. For others, it may have sparked something more and may have gotten them to seek out one-on-one coaching, skills camps, bigger events, etc. etc. By growing the pie, TR (and some of the other platforms) has likely led a lot of people to seek out additional opportunities.
As someone who has chosen to have a private coach, and have found it to be very valuable, I don’t see it as an either/or in terms of TR vs a private coach. In fact, they remain quite complimentary. I still often use TR workouts in my blocks since TR has such a vast workout library its easy to pull something that matches what the coach prescribes. The Workout Creator is so simple and easy to use, it makes creating custom workouts a piece of cake. I TR Calendar and Power Analysis charts remain super easy to use and great for gaining quick insights when I don’t feel like plowing through WKO5. So even though I stopped using TR plans a few years ago, I still find a lot of value in the platform. I also keep a Sufferfest subscription because its fun to throw on one of their video workouts once in a while, I prefer the Half Monty Ramp to the TR ramp, the Full Monty remains a great overall assessment tool, and Abi Carver’s Yoga component is fantastic.
So let’s celebrate more people on bikes getting fit and healthy. Let’s celebrate TR getting more people training consistently and getting faster. And let’s celebrate coaches for allowing individuals to take it to the next level. Keep growing the pie!
there is a massive abyss between general intelligence and focused machine learning
not saying this is gonna be perfect, but comparing “learning training progression rates” to “driving in the world” or general intelligence is not apples to apples here.
Very true, though I would also counter that understanding how the human body responds to training stress in order to optimize said stress is at least as large a problem as driving in many ways. Again, holding judgement till we all see it/use it. Should be interesting at least.
Edit: I, like most, have no idea what I’m talking about.
What it tells me is the extent to which some people feel threatened by what’s coming down the line, facilitated by technological progress, something we’ve seen occur in virtually every other area of life/business through the ages. So if I ran TR I’d take it as a positive signal that other businesses felt somewhat threatened by the direction I was steering my business in…
The good news for these other people is that technology’s encroachment/impact upon their field of expertise will continue to be incremental baby steps, so they’ve time to plan and adjust to the change, as we all have to as the world changes around us. However, they probably shouldn’t linger too much, because per Rudi Dornbusch "In economics, things take longer to happen than you think they will, and then they happen faster than you thought they could.”
Personally, I think it’s good to see former “gatekeepers of knowledge” be disintermediated by technology and progress, helping to further democratise access and open up opportunity for those with shallower pockets. If TR manages to play a role in speeding up that process - and that’s still very much TBC - then all power to their elbow. And if they do, it’s a dead cert that someone else will come along and also steal TR’s lunch one day.
