I’m old enough to remember the old ‘Sweet Spot Base’ type plans, the burnout rate on those was horrific on 3 ‘low volume’ workouts a week. The updates over the past few years have really moved the goalposts, the AI has, so far, been spot on for me. A few workouts look a bit intimidating but have all been completed. For me, the rest is becoming increasingly important, being guilty of adding in additional workouts is a recipe for failure in the long run, the red/yellow seems really accurate. The ability to seamlessly integrate with Zwift is excellent, no more iPad adaptors required. There is also less mental requirement, calendar does the planning, you turn the pedals, end of. Keep up the good work.
16 Likes
I think it’s quite interesting. I have been tweaking about with it, having just flipped over a new AIFTP detection point. I’ve done two workouts, one endurance, one VO2 and both were fine - easy and hard (not very hard) respectively.
So, today I am showing a red day - possibly because I went for a run yesterday after the VO2, but only a short one - tomorrow I have over/unders scheduled. I have some time today, so I looked at moving tomorrow to today, I have no doubt I could complete the workout, probably at ‘hard’ level. But when I do that, it changes the forecast next AIFTP by 10W - which is a big chunk.
Now, I don’t really care about the number particularly, but it is interesting that it calculates moving ONE workout by ONE day as being so detrimental. Overall I feel like I can cope with more load than it prescribes, but I’m trying to exercise self-restraint and follow the programme if I can.
Unfortunately, it’s not always possible due to time constraints and I sometimes end up losing today’s opportunity and then being unable to do tomorrow or the following day. Such is life I guess.
Any difference in predicted failure rate or difficulty?
Had my first vo2 effort after a 4.6% bump
My legs no longer like this update.
17 Likes
Tiny shift only - about 2% more likely to be very hard
I see a lot of die hard TR fans being quite dismissive to other users and their experiences with the new AI features in this thread, and it is disappointing.
Maybe an AI generated prompt with an explanation of the reasoning for an AI FTP bump or drop would be a useful addition to this system and it would significantly reduce the number of ‘TR AI Dropped my FTP by ???’ posts.
What you guys are doing is at the pointy end of the pack, and I understand your primary focus is Road Cycling, and for this group it seems to do it all.
Nate was able to get strength workouts incorporated into the AI model (CEO Perks).
As an Enduro MTB and BMX rider I sit in the middle of these 2 use cases, I do my Strength workouts on the descents and unfortunately for me the AI modeling can’t quantify the non pedaling (pumping, jumping, berms and rough terrain) parts of my rides and (seemingly) fails to incorporate this portion of the effort into the AI analysis, so the model will always be wrong for my use case.
On pretty much all of my descent portions of a ride my HR is flat out but the model appears to see this portion of the ride as recovery as there is very little pedal power output compared to the climbs.
Is there some way to utilize the flow and grit data along with HR data in absence of power output to quantify this portion of the effort as I feel like TR is only looking at half of my ride.
As I don’t have Nate’s CEO powers I can only hope something like this will eventually become part of the model.
2 Likes
What happens when you do not report power data into the model and only heart rate? My thinking here is that might be a better indicator of your total ride / workout.
For those enjoying the TR AI experience so far, what are some the interesting insights that the AI has uncovered for you?
For me, it’s that the TR AI does not like it when I add intensity. When I do so, the TR AI reduces the intensity of following workouts. It’s like it’s trying to protect me from myself!
1 Like
I have done this and I get about a 30%-40% bump in calorie burn for the ride.
This is one of the reasons I took my power meter off my BMX Race Bike, Power alone (when it did work) gave very inaccurate readings for my-ride data. This combined with the power meter missing half the lap because it kept on going into sleep mode while waiting on the gate for the next lap.
Strava AI seems to learn about this activity type but only after me giving it feedback on its wrong assessment and feedback of a 20 lap training night. It has adjusted its feedback to acknowledge the short sprint efforts (10, 20, 40 seconds) with lengthy recovery breaks (1-2 minutes) in future rides.
1 Like
This is the thread for people who are happy. So it kinda makes sense that many in here are dismissive of the criticism that has been abundant on the forum since the new release. These “issues” are at the very least not a big deal, for many of us.
This is my first post in this particular thread, but I have been very happy as well with all the new features. The FTP prediction has let me adjust my training with some changes that in hindsight were sort of obvious - but without it’s influence I may not have actually done anything different and thus had worse results this season.
I had been stacking 3 days together with Intervals-Long Endurance-Intervals because that was the best fit for my schedule. By playing around with the prediction it was obvious the fatigue cost was keeping my final interval session from progressing (and that matched how they had been feeling, my legs were always screaming but I just forced myself to grind them out) so I found a way to add a rest day/recovery ride in between the first two days and along with going down to a balanced training approach from demanding, I got myself an extra 6 watts in my last block and the workouts were less taxing than they were before! Whats not to like.
I actually ride a lot of enduro/downhill myself but I think I take a different approach to my training than you. Part of what pushed me towards structured training was going out on casual gravel rides with some friends who were WAY slower than me on anything gnarly, but on the long sustained flats my legs were burning when they were doing fine. I could punch over climbs and do short sprints all day long, but anything long and sustained destroyed me.
So by training to raise my FTP with tradition “roadie” training, threshold intervals, sweet spot etc I have seen a big difference in my performance for this type of riding. I can do a long climb in an enduro race and I can push the pace a little bit to make sure I’m ahead of the crowd so I can drop in without any traffic. And since I’m not as tired from the climb, I have more energy to push during the actual stages.
I don’t actually have a power meter on my enduro bike, but I did have a fun workout I came up with to do on my XC bike last year. I used a local trail with a ~5 minute climb, and a ~2 minute descent for V02 intervals, dropping into the descent after you’re gassed from the climb is good skills practice for an enduro race scenario where you need to be able to go hard on race stages while relined.
Beyond that, there’s only so much TR could do to account for the exact physical toll your skills work is taking on you. It’s not like you would be doing actual structured workout during the non pedal sections, so all the model will be tracking is the fatigue cost from this. Which using heart rate data does seem to work pretty good. You still have the option of choosing an easier interval workout if you feel like you are not able to handle what the system is giving you. Worst case, you perform poorly/fail some workouts, and the system adjust to ramp you up more slowly and give you easier workouts. That seems like it should workout ok?
6 Likes
@Nate_Pearson stated that heart rate is taken into account and they have avid mountain bikers on the team, including some that compete at the tippy end of pro races.
I would not assume they haven’t taken this into account, I’d presume the opposite.
So to close this one off - I did the workout today after yesterday off, I rated it very hard on the basis that I was struggling today and it has put my AIFTP prediction down by the 10W anyway - which I think is fair actually, late but fair.
1 Like
TrainerRoad AI: Or How I Learned to Quit Worrying and Love the Watts Cannon
1 Like
I have been using Trainer Road for seven years and this is the best version of Trainer Road that I have experienced in that time. My training sessions now feel spot on every time and I am progressing my real world performance.
Previously, I have found that I would make good gains, my FTP would increase and then I would have to push the adaptive training to increase the difficulty to get back to a challenging but manageable level. Now, there is steady progression and I can just do the workout prescribed and know it will be doable and progressive.
That said, I do understand some of the feedback around the change. The new “AI FTP” is a misnomer and I think a lot of the angst could have been avoided if an alternative, more accurate term had been used. I also agree that more user discretion as to whether to enable or disable some of the AI features would have helped to ease people into the new system and avoided some of the anger around being forced to adopt the change.
Overall though, really pleased. Really like the power curve simulations and the workout prescriptions. Great job TR. Glad to read I am not the only one who feels the same.
12 Likes
For me it’s made me think about the order of workouts and that the number yellow/days seems to matter too, in that too many in a period doesn’t help my fitness, which makes sense.
RLGL continues to be really good for me. As opposed to the more formulaic form/freshness calculations out there, it seems much more holistic and often uncanny in its accuracy.
Just now thinking about it, the predicted difficulty is really the thing to look at. It’s essentially the AI “showing its work”. Regardless of what it might have bumped your FTP to, or the type of workout, the %s will show you what it “thinks” the progression is. Going forward if I see an unrealistic change in predicted difficulty (up or down) that’s where I’ll throw the flag for support or the forums.
2 Likes
Chiming in as another satisfied user that is really enjoying the new features. I totally get the the new FTP bumps/drops and projections can be exciting as well as disheartening. I’m taking it all with a grain of salt because after a solid 4 years of training consistently, I know pretty much where my upper limit is with FTP and 5-minute power. I will say it is encouraging when the prediction is above what I’ve been able to do in the past and it’s keeping me way more honest in respecting my easy days and recovery weeks. It’s changing my behaviors because I do want to see those results.
I started a new plan on January 6 when the beta came out and while I was initially put off a massive FTP bump of 267 > 280 (unrealistic) I could make sense how this would translate to the generated workouts due to Nate’s explanations.
Since then, I’ve had one workout that was very hard and that was most likely due to the fact that I was slightly sick. Since then, everything has been moderate or hard. Also, FTP detection dropped me down from 281 > 273 which is spot on.
As someone who has been doing training plans since the Training Bible days, having a structured 3-5 months of workouts has become my comfort zone. My enjoyment of that the structure of that hasn’t changed, but I’ve let go of holding on to the rigidity of doing the exact workouts that have been planned. If a workout changes from day to day, I’ve come to peace with that and it’s one less thing on my plate. The only annoying part is if the workout hasn’t synced to my garmin.
Other features that I enjoy are dynamic endurance, playing around with the pattern of hard/easy days on the calendar, and being able to update the plan when necessary.
So, to the TR crew, I salute you on a job well done.
7 Likes
Despite some early glitches, growing pains and learning curves I am very very happy with it. Serious kudos to the TR team. It was a bold move and it paid off. For me (and all of us in this thread) at least!
5 Likes
I’m not happy with the new AI detection system because it does not take into account that we are not machines we are human with lives and certain things get in the way Work family is more important but it also does not like it if you go out on your bike instead of doing a train session that’s when it wants to decrease your FTP mine is down 4% isn’t TrainerRoad about training to ride your bike
That’s completely fair in my opinion and expected….its like in the gym if you have sporadic routine you should not have gains like with structured program. No system other than Natural intelligence will account for family life and life stress. As parent of toddler I understand what you mean.
2 Likes
When I was at school, I once handed in a poem that I had written instead of my maths homework.
I did not get an ‘A’.
There’s no justice.
18 Likes