So when will this start rolling out? I’m already a TR user.
That 27% improvement is compared to Adaptive Training. There is a break down of the data in the blog post here[click]. Cheers 
1 Like
Same question, when is the rollout?
Adding to what @Brany said for clarity:
Adaptive Training has used machine learning models that we trained to help the system analyze and classify the outcomes of your workouts, and AI FTP Detection has also used models we built from scratch to recognize FTP changes without tests. Additionally, we’ve trained models that have helped give you AI recommended workouts whether via TrainNow, in your training plan, or on Zwift.
But this new update is a complete replacement of Adaptive Training with new models that we’ve trained to do an even better job of giving you the right workout like @Brany said, and new models for AI FTP Detection.
And there’s a whole batch of other exciting updates coming as well
22 Likes
It’s coming soon, and we’ll be sharing more information about the release in the coming days!
17 Likes
Very cool and extremely exciting!
2 Likes
Merry Christmas all!
Am looking forward to trying this out. It makes sense to have an answer to “is it better? How much is it better?” so the 27% thing is a bit of a necessary evil in my opinion. It seems to be described as 27% more often getting the ‘right workout’. Of course that’ll lead to questions like “how is that ‘right workout’ defined? Is there any circularity there?”
Of course there’s no absolute answer though, and the full process used to develop and validate the models would be too complicated (and also probably give too much away) to be useful in marketing material. So I choose to interpret this as “they set a metric to test against and showed a decent improvement relative to that, so it should be fun to try it out!”
2 Likes
Thank you for these improvements. It will be exciting to see how it all comes together. I am glad for the Zwift connection because they have pacing rides and short races which can relieve the monotony of intervals. I’m hoping the new system can measure and account for such pleasures. The last thing I would like to see is Ai train specifically for the “A” event characteristics as a coach familiar with the course would do. The National Gran Fondo Series of courses would be a start and well known gravel events too.
4 Likes
This is great news.
Question: my new plan starts next week. Should I remake the plan or go with what’s already there?
Are there any chances to test it now?
1 Like
There is a thread asking for beta testers . With this announcement the window might be closed
I read that. The chart/table is showing % improvement, but not what the numbers were before the improvement. For example, a 27% improvement on something that’s 20% accurate improves the accuracy to 25%, kinda NBD. A 27% improvement on something that’s 75% accurate gets it up to 95%, so maybe consequential. Merry Christmas!
5 Likes
can we get a tech talk on the modeling techniques used? can’t imagine you’d be giving away any secret sauce without that treasure trove of data
3 Likes
In the next week a podcast episode with Nate will go into more detail.
36 Likes
Congrats team, I’m looking forward to testing it out. Based on my training data the OG SSB HV (and pick your own plan) plan beats adaptive training. This will be another interesting experiment. Fingers crossed for more gains and a better feedback loop.
2 Likes
Looking forward to the upgrade everyone is a snowflake and I believe the AI will do a more advanced interpretation of the TR user abilities and environment to give a better outcome. The old ML model was to me anyway a bit of a task master and adjusting to your own situation was possible I always questioned the decision ( my fault). I am looking to my A race hoping to break the 110km 3hr time. VERY EXCITED 
I have 14 weeks left to prepare…..HURRY UP ALREADY
1 Like
This sounds like it could be a real game changer. The question “what is the best training for me, right now?” can’t be answered very well by other systems or coaches. Potentially exciting stuff.
1 Like
Very exciting. Hoping this 72 year-old can be even more motivated to hit more PRs with this update!
10 Likes
Are workout levels redundant now, and as such, non TR activities will appropriately affect future recommendations?
2 Likes