Just wondering if people have field tested their AI FTP numbers? Post below if so, I don’t think mine is anywhere close to what I could do, and planning to test soon so that I have a pacing number for race season.
3 Likes
Why would you want to do a test that you can fail when AI has given you the correct number? Thats the whole point of TR AI, no need to get an inaccurate result from a test when you already have an accurate one from AI.
This feels like trolling, but I’ll bite.
How do you know it’s accurate without testing it?
There’s plenty of examples in the past year of people testing AI and it being very wrong, and pretty stubborn when confronted with how wrong it is.
If you want to see what TR AI thinks of your ability to ride your FTP go load Cahuilla. It’s 60mins continuously at 99% FTP. It’s predicting a 24% failure rate and I have my FTP manually set 18 watts lower than my last detection. It’d probably be a 50% failure rate without that drop.
6 Likes
How is a one off test more accurate than the millions of ride files the AI has analyzed? TR has far more data than a single test to go off of and get a perfect result.
3 Likes
How are you going to test it?
Hour power? Ramp test? 20 min test (going to do the 5 min test before or skip it?) Something else?
Someone may wish to gauge their performance improvements between set periods. It’s clear TR just made some pretty major changes to how AI determines FTP, which puts any sort of “progress” to past TR FTP’s in question. If someone does, say a 20 minute all out interval session every 3 months, that’d probably be a better measure of performance changes vs a software that may or may not be changing again.
I’m not against many of the changes, but I’m disappointed to have lost a consistent measure of my progress, even if the new program is more accurate.
7 Likes
Perfect result of what? The amount of data doesn’t matter if it’s used or interpreted incorrectly.
2 Likes
I’d probably do a 20 min test and do the fudge factor. It’s always felt most accurate for me and my training. I’m really enjoying the new trainerroad I’m making lots of progress. Just need a true traditional ftp number for pace strategies
2 Likes
Did your FTP change drastically with the new update? How many watts are we talking? I’m guessing it would have to be a significant jump to make a meaningful difference from a pacing perspective? And why would you choose to believe the result of a single effort which will be influenced by your pre-test fuelling, sleep, wellness, motivation etc, all of which can vary wildly from day to day, over an in depth analysis of all of your training results?
Have you listened to Nate’s explanations of the accuracy of the new product on several recent podcasts? If not, it might help uncloud your thinking.
How about using what AI FTP has given you for an event and seeing how your pacing goes compared to that?
Depending on what your event is, you’re probably better off doing a race simulation ride than an FTP test.
If it’s a 6 hour gravel ride, say, it’s probably more valuable for you to go out and see how long you can hold tempo/SS over the course of 4 or 5 hours than it is to have a (potentially) more accurate number to take 80% or 90% of.
Something something performance something something performance itself.
8 Likes
Not since December. I have a long history of ramping after a detection when I think I’m stronger, but my ramp is always with a watt or two. I now just don’t think it’s worth interrupting training, it’s good to get to a point where power/hr is around historic levels but be wary of where you are in your training cycle.
I would instead look at performance baselines, like a timetrial, or threshold reps on your favourite slope.
Which of the many field tests are you referring to?
The previous iteration of AI FTP was meant to predict the outcome of the ramp test. The current version is optimized to keep the Progression Levels in the “center range”.
There is no equivalent field test. In any case, different test protocols will give you numbers that are within +/- 2–3 %. Keep in mind that most power meters have an accuracy of +/- 1–2 %. So a difference of 346 W vs. 342 W is not statistically significant, i. e. since they are within the error bars of one another, you regard them as identical. Even a difference between 350 W and 342 W is a bit above 2 % and hence, just above the accuracy of many power meters.
Beyond that, I’d ask what you want to use your “FTP” for? If it is to select workouts, the current AI FTP works really, really well in my experience. (That’s coming from someone who likes doing ramp tests, i. e. I don’t want to avoid testing.) If it is to pace an effort, I’d say it depends on the effort.
3 Likes
It’s the initial post that’s trolling if you ask me. Just a not-very-subtle way of introducing another ‘I’m not happy with AIFTP’ thread onto the forum. It’s not like there’s a shortage of existing ones to add a comment to.
5 Likes
LLM =\= Machine Learning
Both can be called AI but how they behave is different. So I wouldn’t base any plans to test off stubborn ChatGPT c
3 Likes
I know TR says that now but that is explicitly not what they said when they introduced aiFTP.
1 Like
IDK, but I don’t think there’d be much harm in substituting one of the official TR FTP Test workouts for one of the harder workouts that’s already in a plan. It shouldn’t disrupt the plan much and would provide a trad FTP number. I’m not positive, but From what I think I understand from other recent threads, the new AI would not utilize or be affected by the FTP test result.
I’m definitely not trolling, and I do really like the new training method on TR, as I said. However I think the AI ftp is probably off by 10-15 watts too high. It really does make a difference for me in racing situations to know what I should target for specific sections of the course, breakaways, climbs etc. Like some have said, I do think I’ll need to field test this rather than just use it for pacing with a blind trust. I was just curious if anyone else had done any testing and how it compares, so far not many have said they have, I’ll report back if I do test soon. I really hate testing so will likely procrastinate.
2 Likes
I actually think testing your “AI FTP” / “FTP” is important as the first letter stands for FUNCTIONAL If you can’t express your “xFTP” when out riding, then it isn’t functional, and you might as well call it “TLP”: Training Level Power
1 Like
Yep. Field testing it every workout.
3 Likes