Quite amazing how they specifically addressed some of the concerns in here on the podcast. Any (updated) opinions? 
It’s almost like I knew what was coming… 
Nah, I’m just a TR user 
In all honesty though, if I’m understanding what’s been said, we’ve all been ‘testing’ for a while now. Not that that is really new news.
Again, based on what I’m hearing (and what I understand), this is far more advanced than anything I could have hoped for. Has my opinion change?
HELL YES!
I probably would if I had even a vague idea of what it was 
And he says about the TR release:
Oops! 
Really, mostly this one. He’s not wrong about that.
Add also: Uncertainty about the quality of the ML team.
Quite impressive to see the reach of TR. Seems like some coaches are getting nervous. 
I think the new format will mean you get out what you put in. Using your Carpathian example, if you are delusional and grade the workouts accordingly then your gains are going to be limited. If you are honest about sleep, RPE etc then the customised adaptions will be relevant and mean you see an improvement
Apparently he hasn’t listened to the podcast. Otherwise he would know that only 50% of the user base are following the plans. So self-reinforcement is highly unlikely. Lack of elites also isn’t an issue. > 99.99% of the user base aren’t elites after all.
Coaching credentials seems to be a weird point to state. After all the TR coaches are licensed and some even are former pros. I doubt anyone else got more people faster. But meh the last point seals the deal. As you have stated, nothing more than a gate keeper argument.
Anyway, what coach is working for 20 bucks a month? 
Call me crazy, but before we start poo-pooing the effort in advance, shouldn’t we at least wait how people like it and whether the majority of users feels it is an improvement? And it seems once the plumbing is in place, they can iterate on the algorithm and make sure they optimize the algorithm for the right things.
During the podcast I thought they covered all the important caveats: problematic data sets, the issue of bias, etc.
Check out the TR Bell Curve…”more people faster” doesn’t mean more fast people.
If I want to get fast, I’m going to seek out a coach who produces 5w/kg cyclists vs a service that produces 3w/kg cyclists.
TR produced that 5 wpkg cyclist. 
Anyway, what’s the context. 5 wpkg in your 50s on 3.5 hours per week?
Fast vs faster.
I want to see the coach, who can whip 60+ yo to 5 W/kg
Seriously though, so many of the critics seem to be biased towards the high performance end. (E.g. all the coaches on the social medias).
Virtually 0 of us are pros. I think most TR users just want to get faster withing their means (age, work/life, health, motivation, commitment, goals… you name it). I don’t think a majority even has the desire to get to near pro levels.
So if TR would produce only 5 W/kg athletes, that would only mean the vast majority of customers would be gone. I certainly have no use for a training service, that only caters to the very fast.
Oh and since this is the AI thread, not “the other thread” here a little snippet about ML… I’m no expert, but if it was impossible to apply ML to real-world datasets, it would be a pretty useless technology.
If I’m looking to get faster — 3w/kg — I go with TR.
If I’m looking to get fast — 5w/kg — I go with a real coach.
Nobody is stopping you
That’s a damn big gap between those two points. At what point are you suggesting TR doesn’t work anymore?
Wondering if the AT stuff will shift the bell curve. If it doesn’t, does that mean it worked or didn’t work?
Also, if TR is producing mostly >3w/kg cyclists, the whole POL vs SS argument is pretty pointless because probably ANY type of structured training — cookie cutter or AT — will get most people to that point.
That’s for sure gonna be interesting to see. Would be quite remarkable if it did.