Is this a seismic shift in AT?

[mwglow15]

Yeah I remember the weekly notes being a topic of conversation on the podcast when AT was being introduced. The main point of confusion was that they statically mentioned specific workouts. So once workouts started getting changed automatically it not longer made sense. They briefly talked about making it so the notes programmatically updated with the correct workout names. But I wonder if this just got too complicated as AT progressed since it would be difficult to write the notes in a non specific way to avoid mentioning things like Threshold vs threshold burst, long intervals, short with short rest, etc. but still giving enough context for the flow of the week.

[TooManyDogs]

@Brennus

But, here’s the bottom line, TR observed that very few people actually read the notes. If TR have limited resources with alternate uses…updated notes that nobody reads just doesn’t make it to the top of list.

Forget about the work being done for factoring in unstructured rides and prioritize updated notes

[hpbieker]

@mwglow15

This could be difficult. One of the biggest negatives about these ML models is that it’s just a black box. You put in your inputs and it spits out the output. You can make educated guesses as to why it’s making certain decisions but it’s not easy (maybe impossible) to get a real explanation of why it chooses a specific output.

I do not think TrainerRoad is using (artificial) neural networks for suggesting workouts. My understanding is that they have programmed a way of filtering and ranking workouts for a particular day in the training program, and based on this they will present the best candidate. What they take into account is not that easy to know, but most likely the progression level, duration (e.g. both 105 min and 120 min), workout type including subtype.

So based on this it should be easy to present the reason for the change.

[mwglow15]

Their intro to AT page mentions ML several times and the graphic in the video certainly looks like some sort of Neural Network or graph. That could just be to make it look more futuristic but who knows.

Adaptive Training: The Right Workout. Every Time.

I’m sure they could explain some sort of reason but it might not actually be useful to the athlete and just cause more confusion. Not many people actually understand how ML works (like non tech people) so I think they tend toward less explanation rather than more to not open the flood gates.

[hpbieker]

Based in the video I think it is only clear that they use ML for setting the workout levels. But of course, it is possible they use it for other parts as well.

[Jenalpa13]

Wait a second

TR for years has been recommending training plans that were too difficult???

[Tygerfysh]

Anyone else thinking the big takeaway here is “people aren’t doing recovery rides as recovery rides”? TR is making them 3 to 5% easier across the board `cause they know we all just punch up the intensity of our recovery rides by 3 to 5%. Wala.