Garmin AI – It’s fake
OK, I’ve looked at just one other person’s AI suggestion. Guess what? It’s pretty much identical to mine.
Dcrainmaker shows this in his YouTube video take on Connect+, here is his recommendation alongside mine.
These are effectively glorified templates to show the results of predetermined insights. This was state-of-the-art IT over 30 years ago, and I’m not exaggerating.
I peeked into Garmin Connect’s inner workings a while back, and Garmin certainly does use pre-canned responses in some scenarios.
I’ll give them some credit here. They probably have a broad set of predetermined insight types, stitch together some sentences that match your data, and then run the completed template through AI to change the grammar a bit. That’s not even Machine Learning…I think most people call it Grammarly.
$6.99/month…it’s a bargain.
- More: Garmin connect+ – the RANT Version
- More: Garmin Connect+ – a more serious and sober look at what developments are signalled by the new feature in Connect+ – some strenuous logic even got me to Garmin acquiring Peloton!
PS: I’m tapering, racing, and recovering ATM…I’d normally have higher intensity-minutes 😉
the5krunner.com © 2010-2025
Reader-Powered Content
Thanks for this first feedback.
Let’s remember that we shouldn’t make ourselves slaves to a product either; we can do sport without a wearable after all.
And above all, don’t lock yourself into a single ecosystem, because then you lose the choice and freedom to change.
that is way I use RunGap app and my stats are all over the place:)
and I’m free to choice whatever the watch or bike computer I like
have a grate one
J.
I think this is a real nugget of wisdom. People have become so obsessed with home-brewed metrics that their watch tells them and live by it. Furthermore they will die by the sword defending them online.
I like using HRV4Training because it takes a couple numbers and my input and lets me analyze how I am doing based on a check in of how I feel in the am and what I did yesterday.
In a world of Whoop, Oura, Garmin, and many others telling us how we should feel we lose our abilities to check in with ourselves and learn how we are doing through feeling. I cannot imagine what would happen if Garmin took on RPE in their training plans. People would not know what to do. They are so used to being told how they feel and what they should do.
I guess it distracts from the boring lives many consumers have. These devises and Ai magic are more interesting than their cookie cutter lives. They want to live the life of a super-hero thinking that Garmin Ai or Strava Ai is going to be the thing that replaces hard work and makes them a ultramarathoner.
keeping our sports data device-agnostic means using a platform that is a level of abstraction away from a partiuclar vendor’s native platform
Simpler language – use strava or intervals.icu. 😉
as you’ve done, it’s also a good idea for very specific metircs to use a specific solution – so hrv4t is perfect for that waking monring readiness reading.
i dont both those things (although i prefer golden cheetah and xert as my platforms)
I work in tech and recently worked on implementing product teams desire to provide customers insights based on historical metrics using “AI” which these days is nearly always code for “well just throw it at a generative AI model and ask it for insights”.
The thing is gen AI is basically crap at it. It can produce grammatically correct verbiage but ultimately it doesn’t say anything that insightful, and to avoid hallucinations you have to dial the “temperature” way down so it is all very generic and bland. In the end the tactic that seems best is to use hard data science using established non-generative AI techniques, have those generate a few key canned insights, and then use AI to turn that into human friendly prose linked to an established, human generated domain specific knowledge base.
From the description in the article that seems to be the approach Garmin is using – although I haven’t see the Connect+ offerings myself personally. It’s really nothing revolutionary.
There are some new pre-canned charts and things – I’m guessing those are mostly based on user requests. I’ve asked for features myself – like a chart of your personal best progression. Or they could use one of the tools that marries GenAI to a charting app and lets you ask for ad-hoc charts of any and all your user data. For sure you could do that yourself by dumping all your data and feeding it into ChatGPT – or you know if there was an API that gave historical and real time data that could be opened up to third party developers than I think that would be very popular.
But really as observed in the article