AI Training - Will it work?

So many people think the thing to optimize is FTP or W/kg, but that is far too simple of a definition. Most of us are trying to improve endurance (time to exhaustion, the right side of the power curve) and sprint performance (the left side of the power curve). So really the whole power curve is critical to consider, with some athletes caring relatively more about the left or right side.

Thankfully, TR clearly recognizes this because they are rating your level for different power zones, which is effectively measuring your TTE for different power zones, ie your power curve. And they were clear that different people want to optimize their curve differently and the ML model will optimize what is important to me, not some universal definition of optimal.

Also, Nate mentioned that the ML model is accounting for over 100 factors (features) that affect your training.

So it sure sounds like AT has been properly designed and implemented to really have a meaningful improvement to individualizing the training plans to get the best outcome for each user (assuming you put in the effort of doing the workouts and taking care of your body).

I am super excited for this and I think TR has a huge competitive advantage exactly because of how well they seem to have defined how and what to optimize and guide the athlete to using the tools effectively. I don’t see anybody mailing both of those aspects anytime soon.

4 Likes

Surely it has as much right to be called FTP as any other FTP estimate?

3 Likes

Fitness Assessment
image

2 Likes

Do not try and bend the TSS, that’s impossible. Instead, only try to realize the truth… there is no TSS. Then you’ll see that it is not the TSS that bends, it is only yourself.

8 Likes

I am very curious if this past year, with limited or no racing and considerably more time indoors, offered better than usual data for people who do directly follow plans? I know for myself that I saw better compliance in 2020 than ever before (and not surprisingly some serious improvement in my fitness as a result).

Another question is whether there are gradations to compliance in terms of how TR treats user data? E.g., this past year I largely stuck to mid-volume plans, but during SSB substituted the Wednesday Z2 workouts with the intervals prescribed in the high-volume plans. I also often added time to my Tuesday and Thursday workouts or opted for a + version of the prescribed workouts. Are these patterns something that TR would detect? Or would the changes simply be thought of as ‘non-compliance’?

My point is just that ramp test, 20 min, 8 min tests are indirect measures that estimate FTP.

So output should be called eFTP or something different all together.

Ramp test is a great idea. But TR does themselves a disservice by calling the result FTP - it has caused unnecessary confusion and angst when folks fail workouts trying to be rigid to targets where knowing actual FTP is really really important. Workouts where 1-3% makes a huge difference.

The new progression will hopefully fix this problem. But still an opportunity to alleviate unnecessary confusion by just not calling it ramp test output FTP at all. It is an initial educated guess to set training levels for a 4-6 week block of time and nothing more.

FTP is functional threshold power. If you can’t sit at that power for at least 30 min of time semi-rested then it isn’t functional at all. Same criticism goes for any proxy test.

I’m not knocking the method for initial setting of training levels. I’m knocking the nomenclature.

Here’s a common question: “why isn’t my FTP going up during specialty?”

In an ideal scenario, the answer there is FTP is actually be going up a bit, and duration at FTP is hopefully extending beyond 30 min. It is estimated FTP that isn’t moving.

Again, progression levels ultimately address this point. It is just always confusing to use a different definition of term used commonly in the broader community.

1 Like

*opensDoorAndThrowsIn

what is even FTP? An estimate of LT2?

*leavesThreadGiggling

Yes, same approach seems to work well for me.

What I would be looking for AT to do is make sure that I am progressing at the ideal level for me. I know from personal experience that I can handle some types of intensity better than others. For example, based on the same FTP: threshold intervals and over-unders at threshold I can do without much trouble, likewise short VO2, but give me a set of longer VO2 intervals (3mins plus) and I fall apart. So having different progression levels for the energy systems based on my performance will be really beneficial.

Similarly, would 2 or 3 intensity sessions be better a week. And how much Z2 do I need to make the right progression?

FTP = power you can hold between 30 - 70 min in a rested state coming in. It is both physiological and psychological. That’s where the “functional” part is key.

30 - 70 min is maybe a little loose, but I’m confident on two things:

  1. a nontrivial number of TR users, and cyclists in general, cannot hold their ramp test result power for 30 min in a semi-rested state.

  2. the difference in the power a person can hold for 30 min is pretty close to what one can hold for 45 to 60 min. It is the motivation part that gets hard the longer you go.

Would I rather do a ramp test or 30 min all out effort to set training levels? Ramp test all the way. But it doesn’t do me any good if I jump into too hard a threshold workout because I misunderstand what the score on ramp actually means.

1 Like

An age-old discussion, but I haven’t heard any TR statements that their workouts are indexed against your 30-70 min power. Their workouts are indexed against a Ramp Test result. Because Ramp Test results are just a future workout index, it is irrelevant whether or not it is your actual hour-power.

I agree that calling it something other than FTP would be nice to avoid this entire confusion.

2 Likes

I agree on all that.

It isn’t important that they index against 30-70 min power. It is important that they hit the right workouts at the right time in the least confusing way possible. A really really hard thing to do, and I am confident they are on the path to do it the best way possible.

1 Like

Is this something AI will advise you? I doubt it…
For me it’s still fishing in the same pond but I can be wrong.

I guess we will see. My thinking is that in time the ML working on an ever expanding data set will be able to answer questions such as this. By looking at the training athletes have completed and their results over time, should provide some real insights into whether polarised, pyramidal or a SS only plan are best for a given athlete based on their history and current levels. So advising on overall volume progression as well as balance of intensity. I am not expecting this from day 1, far from it, but think it will come over the next few years.

For now, the use of AT to tailor workouts for specific energy level progressions will be a real step forward in itself.

1 Like

I asked this in the other big thread. Nate responded, but didn’t quite confirm that AT will be able to provide guidance on the best training intensity distribution for an athlete.

Original question + response.

Follow up question.

IMO, getting the right TID is more important than tweaking intensity of individual workouts.

They can’t do that because 90%+ of the data they have is the SS workouts ppl use TR for.

In other words, the adapting plans will look pretty much the same as the current offering, with little tweeks here an there mostly at the workout level…they have no data for the system to recommend anything different.

They do have data (a lot of it) on people who don’t follow TR plans. So it’s not like everyone in the database just does SS plans.

I think the bigger challenge is that its very hard to do. A lot more so than just tweaking intensity up or down.

The biggest issue in my view is the timescale over which to measure and optimize. For TID, you’re talking training blocks or even seasons. It’s not as simple as “easy pass on workout 1” = “make workout 2 harder”.

Edit: I don’t mean to imply that the AT algorithm is this simple, and I’m not diminishing TRs accomplishment with the AT program as it’s been described (I think it’s really cool, and deserves status as a BHAG), but optimizing TID is a much bigger challenge.

They also said last week that they would be introducing a polarised plan (more detail in today’s AACC) and LV+ and MV+ plans. All of these will start to produce useful data, as well as that from those following their own polarised plans using TR workouts.

TID is backwards facing, and shouldn’t really be forwards facing, IMO.

Trying to target TID is teaching to the test rather than focusing on physiological limiters.

Sequencing makes a huge different here; two identical riders can have the exact same TID with radically different sequencing and they’ll have wildly different results.

This is a lot of why the whole “train polarized” thing breaks down for most people.

4 Likes