[Cancelled] - ChatGPT recommended workouts

#1

So we are launching a feature that gives workout of the day from ChatGPT. We can continue to provide ChatGPT more information and better and better requests (prompts) which will give better and better workouts. What we do is search through our DB and find popular workouts that match ChatGPT’s recommendations. This screenshot is not totally accurate but you can see the idea.

image
image591×1280 89.1 KB

ChatGPT tends to recommend hard classic workouts, so we will try to give the option to tone that back a bit. :slight_smile:

Eventually this could be useful, like TrainerRoad’s TrainNow but I don’t see this as a replacement to a dynamic version of Coach Jack. I am looking for beta testers of this. We will launch it in a few weeks. This will only be for paid users.

12 Likes
#2

Interesting, is it just “workout of the day” or does it give you differing types, does it look at recent training ?

#3

Good question :slight_smile: So the cool thing is we can very simply keep feeding in more data and it will get more and more adaptive. To start with it will be fairly basic and focused more on WOD but hopefully in the next month, it will at least look at the last 7 days and give you workouts with more variety, and based on your feed back maybe it can lower and increase intensity.

1 Like
#4

I should say I consider this more of a “toy” meaning just gives you something fun and semi useful rather than part of a real training plan. I can also say that I don’t see most of the other platforms or many coach suggested plans following overly logical progressions so this might be as good as those.

#5

Nice, looking forward to seeing it, the reason that I ask, is that I am trying to not follow a plan at the moment (I suffer from some agoraphobia, and plans give me a reason to stay in), but I do try to keep some structure to my training, I try to do a V02 workout every 10 days, left to my own devices, it will be 7 * 3min every time, having this suggest something close, not outrageously harder , or softer, would be interesting ( 4 * 30 seconds would be silly)

I’m totally onboard with you about the logical progression, I think the idea of every workout needing to be harder than the last ignores real adaption (do you really adapt from one workout in a week, to the point … sorry, you triggered me :slight_smile: )

#6

So Andrea has created word tour winners using progressive adaption but the difference is you start at a very low level and build up slowly so that it never feels super hard until maybe the very end of a training block. It’s a bit more like strength trainers that don’t go to failure but still get stronger.

So I would say it’s not that this is required but slight increases intensity cause you the least amount of damage / i.e. recovery, lowers your risk and keeps you safer/healthier and you are ready sooner to create more adaption. Obviously you are not going to raise your FTP by 50% in one week. It’s going to be lots of small 1% increases… and at the beginning of the season very small amounts of intensity can cause that small X.X% change.

Chad at TrainerRoad uses the term/idea minimum effective dose. Even though TR does not follow this from a tool perspective, this terminology resonates with me. You are only going to increase by a small percent each week and you should do the minimum required to cause that adaption. Andrea, says it does not concept resonate with him :slight_smile: but ultimately that is what his training suggestion is (in my terminology).

Regarding your 7X3 vs 4X30, I would say it depends on the period of your season. So after a base period a 4X30 could be fine and you build to 7X3 for example. Even though you could do more then 4X30 even at the start of the year, it might be enough to cause adaption. But yes if you are in the middle of a training block then dropping a lot likely does not make sense (I never confirmed that with Andrea, but I agree with you).

Sorry to hear about your agoraphobia. When I first moved to Poland I actually would say I was feeling something like that. I never wanted to leave the apartment and the farther I would get from it the more stressed and unhappy I would feel. But that only lasted a few months for me.

1 Like
#7

Totally agree with everything you have said

Yeap that’s what I meant, I wouldn’t want to be doing 7 * 3 all year round

Thank you, I’ve had it most of my life but not really understood what it was, it just got bad 4 years ago when it was actually physically painful to go near my front door, I stopped riding outside completely, and the pandemic didn’t help , cycling is my safe place, again now, but I will ride indoors if I let my brain make excuses, … although it’s not easy, have a councilor now, and she is making me focus on things that I want to do, and has talked me into doing a gravel race in Iceland (wayyyyyy out of my comfort zone on so many level) and did a localish gravel race recently where I came 23rd out of 200 (all ages and I’m 53 … and T1 diabetic) so I’m fairly ( :open_mouth: ) pleased with that

All good now, but I do have to work every at it everyday, which is where something like TR Train now feature works so well for me, as in I don’t have to follow a plan and get it into my head that I must do x workout today, which is better on the turbo, it’s just TR’s progression levels is a bit of a honey trap (for me)

2 Likes
#8

Yes, I agree, the TR train now model is something that many people like and it makes sense. It may not be the best (just looking at complaints on their site) but still it’s what appeals to many and training is as much about doing what you like as to doing what is “perfect,” as what you like might create better adaptions then perfect :slight_smile: No thinking required, just ride and it adjust to me and hopefully the training makes sense at some level.

We want to do adaptive jack, but I think this ChatGTP version could be molded to be something interesting for many people. You can feed ChatGPT “fine tuning” models and improve the results, or at least give them more context. You can also feed in lots of scientific studies and it will start with that as the base to build from.

Yes, I understand this feeling of going near the door. One time my wife was driving away from the apartment and I just jumped out of the car and walked back home it was so intense. Obviously mine is was not so serious like yours but that feeling was so intense even for me. Thankfully, we moved to a better location and I found a new best friend and everything started to change and eventually I was back to my previous self.

Wow, 23rd at 53 and T1 (not sure how much that affects performance). Nice job. I can only dream of something like that. We have a lot of trail races by my house, I am hoping to do some this summer. Good luck on your Iceland race, both getting there and finishing :slight_smile:

2 Likes
#9

Wow, Super keen on this feature & I am more than happy to be beta tester! I feel like this way of training will be great for someone like myself that spends a lot of time traveling around the country for work. Sign me up! :slight_smile:

#10

Super cool. I will let both of you guys know when this is ready.

2 Likes
#11

I always thought your English is too good for a non-native speaker :slight_smile:

#12

I don’t think American’s speak English do they? I feel like I barely can… :slight_smile:

#13

Hi there, where can I read more about the Dynamic Jack and this new feature that’s coming up?
For now I’ve been using Join.cc and TD as a workout player, but my subscription of Join is almost running out and I’m tempted to try TD Jack as a training plan generator.
What I like a lot about join.cc is that you can plan for races (currently only 1 though). But you can create a plan based on an upcoming event. You provide the event date together with the distance and elevation of the event, and then it calculates a plan. Preferably you would be able to enter all your events up front already :smiley:

#14

There is no coach jack dynamic yet. It’s static at the moment but the way it works it easy to modify your original plan builder plans to give in a sense more dynamic but still not dynamic. We also just launched the ability to add multiple events to your plan but I would say it is very beta and does not target specific race categories. Each race does nave a priority A/B/C and has an approximate duration.

As you read here Coach Andrea Morelli is my primary reference source for designing Coach Jack and even though he is 30 year pro-coach he thinks very different than other coaches including the join team. I would say Andrea is more simplistic in nature. We focus on getting you in the right shape for your first A race and then that starts you “race season” where we spend more time focusing on holding on to that peak performance. A new version of our pre-race taper is coming out next week but still it will likely take a few months before we consider multiple races a great feature.

I am very curious about join. They don’t publish much information on how it works and their podcasts are all in Dutch which I only know a few words :slight_smile: We are launching a new product that is dynamic jack with Coach Robert (a human coach) and the beta version of that will be in the next month or so but it will require some heavy Coach Robert in the beginning. Robert is Dutch so he is trying to find more information about how join works for me. Mostly just out of curiosity, not trying to copy them as I can see fundamentally we think differently.

I guess you being a join user understand it :slight_smile:

2 Likes
#15

Hi, thanks for the elaborate reply!
One of the neat features join has, is that it is very dynamic in its 10-day schedule. You are also not able to look further in the future to see what it has planned for you.
So basically every week you provide it with you availability, and then it starts calculating. If you on a day don’t have time, you can mark a workout as skipped (for whatever reason), but the plan will then automatically recalculate the coming workouts to make up for the missed workout. Also if you prefer to ride outside, and not follow the workout 100%, it will take up the workout based on strava data in its plan, and will update the coming workouts taking into account the stress from that training. It will also take up any other strava activities and estimate the amount of training stress it took, for instance if you went for a run, it will also recalculate your upcoming trainings based on the intensity of that run.

In that sense, its easy for a user to not really have to shuffle any workouts manually, as it will always recalculate for you if you missed a workout or did something else.

I have now played around with the Jack workout editor for a bit. I am actually suprised of the many features already existing! I will definitely give it a try! Only thing I’m a bit stuck on is with off-days, for instance I know I will not be able to do a long workout in a specific week, I see I can add it as an off-day, but after doing that I do not see the other workouts around the off-day changing to more ‘stressfull’ workouts.

Anyways, if you have more questions you can always send me a message! :slight_smile:

#16

Sorry for the slow reply. Yes, Jack is not perfect for days off. We really need the dynamic version. I will likely ping you on your thoughts ideas.

1 Like
#17

So this ChatGPT is going slower than I hoped. I thought this would be easy. Well it is easy but it is not easy to get it to product what we consider good and interesting workout combinations. It tends to recommend the same thing every day. This is more a problem of our prompt design as it’s obvious in testing that chatGPT can do something interesting here. Also we want to make it fast/faster.

3 Likes
#18

A post was split to a new topic: Improve Garmin Intervals

#19

Happy to be involved in beta testing

1 Like
#20

Sorry guys we tried to do this and it seems we could not get GPT to give us good recommendations. Now after a deeper understanding of GPT it’s not likely the best solution for this. GPT is best as an assistant (natural language processing) or a tool that can consolidate the most popular advice from the internet. You can build your own rules engine the GPT connects to but then ultimately it’s is just your rules. There are other emerging AI technologies for this but we decided to just launch the TODAY feature instead for now, which is already done.

3 Likes