My feedback after training with AI workout

Hi Scott,
I didn’t think about trying that. I will give it a shot next time, it sounds promising.
My far bigger feature wish would anyway be to see low/high/peak XSS live on Garmin during the ride :slight_smile:
Thank you, apreciate your help, as always!

2 Likes

I am not sure what benefit you would get from seeing high peak XSS during a ride when doing a workout.
Surely the workout states how much of each strain you should be getting when doing the ride.

I am refering to riding outside, no workout. And the benefit would be to see how much you have done and when you hit your target for the day, without overdoing it.

3 Likes

I haven’t had time to figure out the new features, just learning about the Autogen stuff tonight. I can’t get the option to work, I have no autogen button, I started a separate thread on that but happy to get help here. After looking at it though, I think I’d rather just have the 3 totals on my screen and let me free ride it that way. That would be so much easier to do than doing a structured workout outside where the terrain at the moment doesn’t fit the prescribed interval. It’s basically what riding to focus was before but more precise with the three XSS numbers on the screen.

I agree that a new Garmin data field with the XSS breakdown should be added. It would be useful when free riding when trying to match today’s XSS deficit. I guess this is not on the top of their list and that code changes for the Garmin will require different skills then what they typically utilize in the rest of their system.

1 Like

It’s a neat idea, for sure!

The challenge is that showing the numbers doesn’t really instruct users how to achieve the needed L/H/P XSS. In fact, that’s exactly the point of Focus/Specificity - the distribution of L/H/P is used to calculate the Focus duration & specificity, which gives user’s actionable info for their training ride.

For example, if the goal is 123 ( 115 | 7.5 | 0.7 ), the next question one might ask is 'How should I ride to get exactly 115/7.5/0.7? In the past, we had the Focus Duration (in the example above, Focus Duration: 8:28), but some users still find the concept of Focus too abstract/confusing.

The new AI-generated workout aims to eliminate that. Rather than giving you a Focus duration, we can just give you the workout that meets the recommendations!

We also have some simple updates that will arrive soon that also aim to explain Focus/Specificity in easier terms. People may not understand what “Mixed Breakaway Specialist” ride means, but maybe it becomes more understandable with simpler terms:

  • “Mixed Breakaway Specialist” becomes ‘Focus on attacking with strong power efforts with medium rest intervals between’
1 Like

My issue with focus always was that adding any extra endurance to the prescribed ride/workout would tune focus down, usually towards “something-endurance”.

Now you might say, that I just should not add any extra endurance. After all I set my rider type intentionally to “climber/GC/whatever” rather than “endurance”
well, but then I think my training load would significantly shrink. Just because I can’t sustain sufficent intervall training to compensate for the volume I could generate with (extra) endurance riding.
So in that scenario, focus didn’t really do anything for me.

On the other hand, if I would know I need to get 3 high-XSS at my 10min power (=climber) in and I’d see that high-XSS live on the garmin, I would know exactly what to do, no matter how much endurance I do before, after or inbetween generating that high-XSS.

If you could resolve that knot in my head, that would be great!

2 Likes

You are correct though you can just get 3 high XSS on it’s own. It comes with low and peak as well hence a “focus” might make the concept easier to grasp. In the end though, if you want to raise your HIE and you need 3 high XSS for that, once you get to 3 high XSS, you can ride all you want below threshold (lengthen the focus). You’ll still have 3 high XSS at the end of the ride.

I think that is why we would like to see a new data field before the next outdoor season. Be able to see that we have done all the hard work before we just add some easy volune at the end.

Let’s say you plan to do 9 x 2 min @ 115% as part of a longer ride to produce the wanted high XSS. You don’t run it as a workout because you want to do them when it fits the terrain. Some of these will be short of 2 min, some longer, some at a different power. If you have the extended XSS field you can continue with the intervals until you reach the XSS target you set without following a strict workout.

3 Likes

I agree and we have to consider this as an important addition to the Garmin datafields but also to our other real-time software applications - EBC and Karoo integration - as well.

These missing elements are why this is called an interim release. There are many additional capabilities we’ll need to add to the apps and datafields, as well as other improvements to the web app too that are needed. This will become more apparent as some of the new features get announced.

But all this is simply too useful to hold back until we have everything completed so hopefully that’s ok by the user community for now.

3 Likes

I’m not sure how a high strain data field would be actionable as you and Scott have pointed out - you can’t target just that number since efforts above threshold also generate peak strain, and the resulting focus (mix of low high peak strain together) may not be ‘right’ depending how you try and do it (and how you’ve generated high strain already in the ride)… not sure if you have in mind an alternative data field / approach to help?

Another question is whether updates will clarify whether the intention of ‘recommended focus’ during different training phases is meant to apply to your whole weekly training load, or just on intensity days? Depending on how many intensity days you have (and how long they are) the resulting weekly focus can be hugely different. Said another way, if you want a weekly Rouleur focus and do two endurance rides a week, the intensity days probably need to be puncheur or shorter focus to achieve that over the week.

And with the changes to training status, it seems the resulting polarization is different e.g it’s been mentioned elsewhere that those with lower TL will see less yellow (so more intensity) than before while those with high TL will the opposite (so less intensity and presumably ‘longer’ weekly focus durations than before…) This is linked to comments above where you suggested that once you hit a ‘target’ high train you can just ride endurance, which lowers ride (and weekly) focus

I think it is. Thats even how xert is build:

  1. You have a current high training load of e.g., 2.0 and some positive improvement rate
  2. You have set an athlete type, lets say 10min (climber)
  3. XATA wants you to increase that high training load to make you climb faster. So it prescibes a bit of high XSS.
  4. You go out, riding however you prefer. But among that, you target to generate the prescribed high XSS riding close to your 10min power.
  5. Job done. High training load has increased, while you adheared to specificity. xata will continue to push you further next time.
  • Yes, shorter power athelete types will also want to explicitly fullfil their peak XSS targets of course.
  • Polarity can be kept high or low according to individial preference. Just stay below TP during your ride apart from that surges at 10min power if you prefer to have it. I myself don’t really care and just prefer following terrain and my feeling on that day.

So that is very actionable and easy to do. In fact the new AI workouts work exactly like the above as well. They translate target low/high/peak xss to one specific layout of power over time (among unlimited possible variations). The result is just a bit less flexible than doing that freely by yourself according to upcoming terrain.

This I understand, and agree it’s a way to manage outdoor rides, but that’s riding to focus, not riding to hit a high strain target. And it’s also not ‘adjusting your outdoor ride based on whether you have hit a high strain on a data screen’ (paraphrasing a bit :slight_smile:). To me, seeing a high strain number mid ride, and even having a high strain target in isolation (independent of low and peak strain) don’t really help. It’s the focus and XSS that counts in Xert world (or at least has been till now)…

I think what you’re saying is that combined, riding to focus, plus a volume metric (high and peak strain) could be used to guide how much intensity is ‘enough’ for a day. That makes sense. A related question though is if you have more time, should you just add endurance (and lower the resulting focus) or also add more high and peak strain to maintain the focus. Am not sure there has been a clear answer to that so far…

After being out sick a week I finally felt good enough for some HI work today and gave Scotts “flexible AI workout” approach a try.

  1. I generated an AI workout today:

Its stats pretty much matched the xata recommendation
image

As I didn’t want to go full in right away and in general would prefer to spread HI efforts a bit out over the week, I decided to only go for 2/3 of the indictated sets, leaving the rest for later days.

Question: Would at make sense to have an option to advice xert to spread HI efforts out versus prescribing all for the same day? I thought the polarization slider setting maybe would do that, but it doesn’t seem so. Mine is set complete left (“none”).

  1. Setting up the workout on my Edge
    In the generated AI workout, effort power slightly differs between sets (higher->medium->lower). I just used the power of the medium set for all of them to avoid a more complicated workout structure and having to update more values.
    So, as I already had created a workout with the typical AI workout structure on my Edge, I just needed to update the 3 key values as prescribed for today (effort power, effort length & #repeats per set). That literally took me 20s.

  2. Going to ride
    All worked as expected. Loaded the workout on the edge, started the activity. The warmup interval went by and the Edge waited patiently for me to press the lap key to start the first HI set. Which I did after reaching my fire road climb being warmed up.
    I executed the repeats as prescribed. Some of them I pulled together, jumping the rest interval via the lap key. Some sections of my climb are just too steep for having any meaningful “rest”.
    Else wise, during rest, I just kept climbing in the first gear with minimum possible power. I didn’t even program any number for them into the workout.
    Took the trail back home as happy end…

And here the result:

  1. Review

You see, the achieved focus was classified as “Mixed Breakaway Specialist” (5min). Compared to the goal set to “GC Specialist” (8min).
I assume that is because I overachieved most of my efforts in terms of power target. It’s just too hard to match 8min power on a variable climb. And I guess I am more the better-too-much-than-too-little kind of person.
And I also decided to not care about being too precise with training prescriptions. Until someone starts paying me, I just bike for fun.

Xert prescribed me 5.2 high XSS, following the xata recommendation.
I targeted doing about 2/3 of that, but looking at the activity in the planner, I overachieved:
image

Also fine.

Summery:

  • Manually replicating the AI workout structure in Garmin works well and allows more flexibility riding outside, by allowing to start work sets on demand via the lap button
  • Once an AI workout is created on the garmin device, updating it to match the demands of a specific daily AI workout is fast and easy
  • With that, the new AI workout function should allow me to steer my progressive overload much more precisely riding outside.
2 Likes

out of curiosity: the manually replicating, do you then:

  • make a manual workout in garmin connect
  • make it in a 3rd party like intervals.icu that then syncs it to GC
  • i dont think native xert WO builder has the button press to end, or does it?

nice results on the test btw

I tried to build a workout on the Garmin Connect web page, but that seems very poor and missing needed options.
I did it directly on the Edge device’s integrated workout builder. That worked great.

1 Like

@ManofSteele
Why do I see the progression improvement in high training load with 1d delay? So for the 24.01. in example below, while the activity was executed yesterday, the 23.01.

That upwards incline caused by my yesterdays workout only showed up today. (I didn’t do any activity today yet)

interesting, I used the AI generator and I got exactly the same workout as @Beutelfuchs shows above:


so from a technical point of view why would the AI generate the same workout for two different individuals? Sound like the generator picks from a selection of workouts it has in the can and says this is a good option for what is your situation? Is it really AI generation or AI selection?
I ran it a few minutes ago and lo and behold look what it advised:
image
look familiar?
In fact I have run the AI gen a few times and this is the only workout it delivers.

I know Xert put AI in the title. I would say AI is not a good term (not only for Xert), they better use Auto generated, Optimized or Personalized — or something similar.

So basically they have a few workout variants (or only one) where they can adjust a few parameters like the duration of the intervals, number of intervals and target power. Then they try to find the workout that will give the highest score, that is the best match to XSS (low, high, peak) and difficulty. And maybe they take the time into account in a way.

But it is not magic. It will not invent a totally new workout structure that is better for you based on your previous training history. It is basically generating all these variants in the background to see which one will be the one to show at the top of the recommendation list for you right now.

1 Like

It does help when you read the instructions and help tips. Or am I the outlier? :man_shrugging:
Autogen Workout – Xert (baronbiosys.com)

Example of XATA and associated AutoGen workout –
XATAnow

Could you arrive at similar strain numbers with more variation in interval types?
No doubt, but this is the first iteration of Autogen. It’s designed to create a “simple workout” for a reason. :wink: Gets the job done efficiently. Works indoors or outdoors.
If I don’t especially like it, there is a list of alternates I can chose from.