My feedback after training with AI workout

I can see that the AI gen so to speak is not that AI, it just does an optimization using the simplest algorithm to get the 3 XSS scores close to the optimal for your workout. It would be good if you could specify short or long intervals then it could tweak the intervals to hit the optimal scores. It is not that hard to do manually. My suggested work out was close but not right on so I just went in and changed the high 9sprint) sections and the moderate longer duration segments to get it to the optimal or much closer to the required settings. If you only do short intervals per the AI generated work out then you don’t hit the longer duration systems in your physiology. So that part seems to not be to “intelligent”. I would like to see it look at things a bit more wholistically to say you need more long intervals to improve your aerobic systems vs your anaerobic systems and then generate a good workout for that. I don’t think the algorithm is that difficult but I am not into coding any more so maybe it is harder than I think. If the system looked at say your last workout and determined you did a lot of long threshold intervals then maybe this time it should be short intervals and vice versa? The one good thing about the AI gen is it made me look at the XSS breakdown for what I should try to accomplish and then I figured out how to tweak a workout recommendation to get closer to optimal. Not a huge thing for me maybe a huge thing for a math engine.

2 Likes

I find the z2/z3 generated trainings very time efficient which I like, and have experienced the Tempo intervals ranging from 10-15 mins which has felt perfect so far. I agree that when intensity has been asked for the targeted workout session, the AI generated intervals have been of short duration around 1min). I don’t think that is too big an issue (in my limited experience) as there seems to be standard workouts ‘close-enough’ to what is required. Would be interesting if the auto generated workout durations increase/decrease in duration depending upon what the current training focus is. Overall, I’m finding that the auto generated workouts have been spot on for me and very efficient!

1 Like

Please keep in mind that these target outside riding. I’d feel the moment you enlarge the interval time significantly, first riders would complain that their terrain doesn’t allow to safely complete them (short climbs, road intersections etc).
On the other hand, you can easily have longer intervals, just jump over as many of the rest intervals withhin sets as you want → lap key on Garmin.

1 Like

that may be another point of variation, is it for indoor trainer workout or outside? I agree and actually have lots of room to do outdoor intervals should I want but I do not think it is worthwhile vs what can be accomplished in doors. Also if you skip rest intervals in a set training piece it will change the 3 XSS ratios set in the workout and change the focus. So I prefer to be able to set and play for workouts on the trainer.

One of the biggest reasons for breaking intervals down into short(er) duration efforts is to keep the difficulty score low. Extending intervals to be longer in duration can be tricky, especially for higher intensity efforts, say at efforts above your ~5 min power. This is because XSS rate for a constant power interval will increase as MPA declines, so the longer the intervals become, the fewer of them are needed to hit a strain goal. However, because the difficulty score/rating is tied into XSS rate, you’ll also see difficulty score increase quite a bit as you start extending intervals.

To give a comparison, I had Xert generate a Pure, GC Specialist focus (8:00 power) and then modified the interval duration & # of repetitions to keep the total work approximately the same. Unsurprisingly, essentially doubling the ‘ON’ intervals led to a little more XSS with similar amounts of work performed. However, the modified workout had a DS of 90, compared to 77 for the original workout. The last interval of the first set would require me to dig pretty deep, tbh:

I also had a thought… the difficulty score reaches a higher level much sooner in the longer interval version. If you were to integrate the difficulty score over time, you could compare workouts not only by the highest difficulty score, but the cumulative difficulty, if that makes sense (?)

For example, the max difficulty score of the original workout was 77. The modified workout hit a difficulty score of 77 by minute 34 and reached it’s max rating of 90 at 20 min later… That’s 20 min of riding at a DS higher than the maximum point from the other workout.

The hard part is getting them all to align towards a specific outcome. It’s not a single lever or double lever that you need to move simulaneously to get the 3 value target while minimizing duration and difficulty. We called it an “AI” because the problem really can’t be solved manually by a human, at least not within seconds. You ask it to give you a workout that generates a specific beneift and it finds one that you can do. Creating a workout yourself, you can probably get some numbers right but you’ll likely be playing whack-a-mole with power and duration targets since there’s no easy procedure you can follow to solve for a solution. If there was, an answer would come back pretty quickly.

I think there are improvements we could make and I have the idea that the workout definition could be optimized in real-time as you perform it to help you achieve the outcome. Sort of next-level smart workouts. It could also take duration/difficulty inputs in real-time to shorten/lengthen, make easier/harder as you perform the workout.

One of the cool things about Xert is that you can do the same workout inside and then outside, and see how different they really are, in Focus, Difficulty, and overall XSS

At least for me, they’re not that different. I find the TSS just about equals out. Specificity tends more toward Polar outside, slightly.

I’ve always loved that I can modify the length of intervals, where intervals or rest just don’t suit the terrain, and still get the Focus/Diff/Specificity that I want.

Because I was still thinking about it, I built this alternate workout to the classic 3x20 at TP.

Classic 3x20 → Xert - Login

Workout with 7 minute efforts, and 4 minute ‘rests’, to accomodate a hill I have. → Xert - Login

As you can see, they’re almost the same in time, Diff, Spec, XSS, and Focus, but wildly different in structure. Xert is cool for this.