And a number of them are using FTP values from months or years ago. If you are doing level 1 Threshold work and it’s very hard, perhaps it’s not your FTP.
I’m just grumpy as I just had my second surgery in the last 10 days today. Too much walking and not enough cycling.
TR should just ban these people, complaining adds nothing to the conversation and if they are unhappy with TR why bother being on the TR forum?
It’s one thing to go in with curiosity with some frustration vs angry and closed minded. AI is probably part of it. Although the technology was being used previously.
A change in the prediction of less than 1% and people are losing their minds.
For me the key point in the original post is without explainability. I hope that in the future there will be a paragraph to go with the FTP prediction, or just after a workout that gives you something like the feedback a coach would give. It would be great if it said “You struggled in the last interval where actually we expected you to finish strong. Based on this your progress will be slightly slower, but keep at it”
Without this there is no context to why the prediction has changed
The lack of explanation is sort of the nature of the beast with a lot of AI.
If you could explain why it’s making the decisions then you probably wouldn’t have needed to use AI in the first place.
I think they are a long way off from being able to give simple and accurate explanations - but they may be able to add some rough guidlines?
Aye, but its science - evolution - improvements in understanding and application to make it better. Advances.
I mean the advertising banner at the top of the forum literally says “want to raise your FTP?”
TR is still very much focused on FTP.
90% seems low. 
This x1000.
AI tools are great for taking huge data sets and finding trends/relationships that a human would struggle with. So, it can look at everything you are doing with your training, and then make a forward projection based on historical trends/results for athletes that followed a similar pattern. Why did the projection end up where it did? Because that’s the most likely outcome based on trying to match you up with known trends, similar athlete profile, etc. And this “pattern matching” approach can sometimes create outcomes that don’t make much sense. You might absolutely nail today’s workout, but that “good” result might shift your pattern enough that you end up in a group that is trending down instead of up. The explanation would be “because’”, nothing rational other than that’s what the data predicts.
A machine-learning AI system doesn’t have a lot of “IF Then” logic baked into it. It’s not giving you “credit” for successful workouts and “debits” for hiccups in your training. It’s likely just trying it’s best to match you up with trends it’s identified and using that to predict where you’ll end up. It’s like when you have a bunch of possible paths for a hurricane, the percentage likelihood of different outcomes is changing by the second. Today, you might have a 75% change of having a 300w FTP in a couple weeks and a 25% chance of having a 307w FTP in a couple weeks. Something seemly meaningless overnight could swap the percentages just because it’s recognizing a different pattern. If you are following the plan to the exact wattage/time in the workouts (hard to do exactly), I wouldn’t expect to see things shift too much, but keep in mind the the patterns you are being matched up to are constantly evolving also (improving in theory).
The explain-ability is exactly it for me. I couldn’t care less about FTP. What I do care about is a plan with a schedule. I pay a premium for TR and the one quantification I have that the training is working is the TR version of FTP. Progression levels are fine, but they reset. Power curves are only for all out efforts for a given time interval. I want the system to explain pretty much every change it makes. A good portion of people that train with intervals, plans, and science are analytical. Are my workouts properly scaled? I don’t know. I never fail any anymore, any 99% I pick the RPE rating it expects. But that’s exactly the point, then why do I never hit the estimate?
Finally the forum gatekeeping that you can’t ask questions and express concerns about AIFTP or FTP in general without a bunch of bashing and name calling is awful. Calling people crybabies and egotistical in this thread is childish and counterproductive and it’s a pattern. Excessive gatekeeping is the Canary.in the coal mine of forums. If you don’t like the topics, maybe just skip them rather than pipe in on each one to be prickly.
Yes. Agree. This is why I’d rather have an algorithm based system over an AI one. Then a normal human can back calculate the results and make sure the AI isn’t imagining things.
Unless things have changed, Join also changes your workouts and doesn’t explain why.
Looking at your calendar, a lot of your rides have power output that is VERY erratic. I’d guess it’s possible this is getting in the way of your planned progress. Have you tried reaching out to support to see why they think you’re not seeing performance improvements (not sure if you meant just FTP or points on your power curve).
No, this is
This sounds more pesimistic than it needs to be. I’am an academic researcher who (also) works in the XAI (explainable AI) area. In many cases we can already show why the model under consideration gives you the predictions you get. There is quite a bit of research in this area. AFAIK TR uses a transfomer-based model, and there are quite a few techniques which may help to explain the way the model works.
I did exactly that. 4 weeks where I followed the plan TO. THE. LETTER. Didn’t fail any workouts. Didn’t skip any workouts. Didn’t do any extra workouts. Rated endurance rides as ‘easy’ VO2 as ‘hard’ and SS as ‘moderate’ as I have been doing since those ratings were introduced.
Initial prediction was +9W which decreased about 3W each week and now, with 2 workouts left, prediction is +0 W (no change). I see a couple of explanations, none of which are very helpful.
I do acknowledge that FTP (much less an AI predicted FTP) is not the be-all end-all of cycling performance. But we gotta measure SOMETHING to figure out if those hours sitting on a bike are having the desired effect or just wasting time.
Have you reached out to support? What did they say? Your calendar is private, but if you wanted other users to have a look, you could share screenshots of the 4 weeks from detection to prediction.
Were the workouts on a trainer in erg, trainer in resistance or outside?
Were the workouts that you ended up doing the same workouts that were assigned at the start of the block?
The thing is, two riders can do exactly the same training and see different outcomes. There is no guarantee about outcomes.