Stumbled across this on instagram. Excuse the crude phone screenshots.
Caution:
Any software that says itās using AI - but then refers to rules in its description of how it works - isnāt using AI or machine learning (at least not anything thatās any good).
Context:
I spent 10 years building Ai/ML models, and now I advise companies on how to (and when not to) build them.
Iām skeptical of that ad.
@Nate_Pearson and team have been pretty transparent about their approach, and in general it uses real outcomes (perceptual survey results, workout success, and improvement in ramp assessment results) for training and applying their models.
3 Likes
Agree with wake, and add to it if they have no way to have a dataset of substantial size to teach a model, they dont have ML. Dont have something like millions+ or of rows of information, you either dont have ML or have poorly trained ML.
3 Likes
Well, technically a rule based system is AI. Iāve seen rule based AI beat complex neural nets at some tasks, it all depends on your training set.
I completely agree with you that rule-based systems can outperform AI and ML-based models.
Read anything like āSimple Heuristics that make us smartā
But a rule-based system is a simple decision tree.
Itās not machine learning or artificial intelligence.
Rule -based systems and AI/ML are two different ways of arriving at conceptually similar outputs/outcomes.
2 Likes
Rules or features? Maybe Zihi dumbed down the description.
Nate said the TR ML has over a hundred features to define training and how it should go. The mountain of TR data is based on TR plans vs actual.
Which comes back to the TR plans. My perception of TRās overall philosophy - a time-crunched belief that progressing power zones is better over the modern-classic approach to aerobic engine building. And the TR approach didnāt work for me. Before AT, the TR plans appear to focus on quickly progressing difficulty of workouts in specific power zones. Adaptive training has personalized the ramp rates, but the basic philosophy remains. Which is great if that works for you.
Ultimately I improved ability to do VO2max work by doing mostly endurance work on a budget of 8 hours/week. Focusing on the metabolic fitness side of the aerobic engine, and not cardio fitness. Boring old endurance+intensity work (not long slow distance, but what I call the modern-classic approach to aerobic engine building on a āmediumā time budget). That is a very different approach versus having ML pick a lower difficulty vo2max workout starting point, and progressing from that lower level.
Iām curious about a system with over 900 coaching rules that purports to identify training opportunities and craft a custom plan. Because there is more than one way to get fast.
Expert systems count as AI
No where near the complexity of ML systems so canāt really see trends that arenāt already well known. They donāt really get trained so much as finding bugs in the decision tree of the system
1 Like
Showed up in my facebook feed. Not too subtle.
Pic removed
Imagine Zihi and a lightly clothed female advertising āAdaptive Training Plans.ā
1 Like
Zihi was posted last week:
Thanks. Looks like the pictures are getting āworse.ā
the one above looks like a slightly racier version of spin classes at our gym. Iāve seen plenty of cleavage, but whats up with those pink shorts? Reminds me of when my daughter was in club volleyball 
Do you mind nuking that pic and just stick to the existing thread with proper discussion?
Mid SS workout and Iām not able to handle this at the time.
1 Like
FWIW I believe the pic is not offensive or obscene and speaks to the marketing strategy. I get your point out and f moving it to the original thread.
Anyone know if TR has a trademark on the term āAdaptive Trainingā? Seems like the Zihi use of it is directly competitive.
Far from conclusive, but there are no signs of TM or R registration (which would usually follow the terms) on their main marketing landing page:
ETA: Here is all I could find:
https://trademarks.justia.com/owners/trainer-road-llc-2986173/
hmm. wierd. I think if I were them Iād be looking to protect this terminology since itās intended to be a differentiator for them.
Could you share how that 8 hours broke down into workouts?
I would be interested to see since I found TR plans very tiring. TBF, I havenāt tried adaptive training since, as you say, the fundamental approach hasnāt really changed.
On top of that, a quick G search shows that there are likely a number of places with that term/phrase in use for a while.
The team page on Zihi site has this:
80% - aerobic workouts and 20% - high intensity
which is roughly what Iām doing, although when we get into builds its more like 70/30 or 60/40.
Something like this right now:
- Mon 1.5 hours with some intervals
- Tue 2 hours aerobic / endurance
- Wed 1.5 to 2 hour group ride
- Thur off the bike with 50 minutes core (glutes/hamstrings/posterior chain) work & upper body strength maintenance
- Fri 1 hour intervals
- Sat 2-4 hours aerobic / endurance / group ride
- Sun off the bike with 50 minutes core & strength maintenance
thatās a little more than 8 hours, and the average is brought down by ālife happensā missed workouts, and rest weeks in the 5-6 hour range.
Averages and weekly totals:
Upper chart has dashed lines at 6 hours/week and 8 hours/week.
Weekly average is 7:37 hours/week for the year, and just looking at it you can see there is a lot of variance because of life happens conflicts and my usual summer off-season.
Heart rate polarized breakdown is 76% low intensity, 19% mid intensity, and 5% high intensity. The power breakdown on intervals.icu is 74% / 22% / 4%. Same same from a big picture point-of-view.
its not just that, because common words or phrases can be an issue. For example apples have nothing to do with computers, so AppleĀ® can have a registered trademark for computers, computer software, computer peripherals, etc.
Just like Scholastic publishing company has a trademark for Apple, as does Apple Rubber Products.
So the word Apple can be trademarked by several companies, provided the company business is unrelated to apples, and each company is in a separate vertical / unrelated business.
Which means if you sell apples then forget about asking the USPTO for a trademark to Apple. Some info here: Strong trademarks | USPTO if you are interested.
Anyways, back to Zihi. Started as an idea in 2012. They received venture funding in 2020 and hired a team. In 2022 they released support for XC skiing training plans based on coaching rules/patterns including the Norwegian polarized model. Next up are plans for cycling and triathlon.
And interesting marketing FB strategy, the removed picture had a woman in a spin class. That was the first thing that caught my eye, and my brain was picking up on some subliminal messaging. So I looked closer and she was wearing a short pink tennis skirt and tank top. Now it is not an uncommon sight at our gym to see women not playing tennis but working out in short skirts, and yes, occasionally one is seen in the spin room.
1 Like