i’m 3 days into my trial… and i’m wondering if my experience so far is typical. I love the idea and the features available, just the delivery isn’t quite working.
I chat with AI about my training approach and the plan it set. I’m mainly training base/LT1 right now - so want to remove any tempo runs midweek (as a random example). It suggests the plan is updated, so i say go ahead.
Sometimes it thinks for a while, then just gives up and respsonding at all.
Other times it appears to delete activities (maybe a few at a time), and prompts. If i’m lucky it then adds replacement activities and prompts to accept them.
One time it seemed to give up and simply had just deleted 2 activities.
Additionally when it prompts about changes - you don’t have much context , not even the date its operating on. So its really click “keep” and hope. As a site point - i’d argue “keep” or “skip” is a strange wording for deleting an activity. As a 2nd side point - lets delete and add each date in one shot?
To add to this its all painfully slow - like waiting 10s of seconds betwen each prompt, or minutes until its given up.
Compare to using striaght ChatGPT its incredibly frustrating. Yet its ability to set a plan and sync Strava/Garmin runs is unmatched and potentially brilliant.
Also when i switched to Gemini Pro, it seems minutes to reply - so i gave up on that!
Just fixed the issue with the date not displaying, you should always know which date it wants to skip from now on. Thanks for pointing that out.
Re speed: It highly depends on what you’re trying to do. On every chat message you send we have a classifier running that decides which tools it needs to address your request. The more tools it loads into the context, the slower. Workout creation for example can be highly complex so once it decides it needs all the tools, things tend to get slow. That classifier usually takes less than a second but at certain times can also be a bit slower, up to a few seconds.
Gemini (same with Claude or GPT 4.x) through the API is also not as fast as using Gemini, Claude or ChatGPT apps and speed varies. In general, those guys keep the real speedy inference for themselves
Will have another go at improving the chat speed as soon as I can
thanks, i see the date added .. i’ll keep testing. I’m wondering about how bulk updates might work - am i’m meant to replan via the plan settings or via the chat? The chat seems the best way to adjust - but right now would prompt way too many times if i needed to do a rework due to illness and injury resetting my volume.
unless you want a whole new plan due to overall volume/availability changes, or re-organize a particular week, at the moment I would use the chat. We’ll also have something to bring you back to compliance if you’re not really following the plan similar to the “Based on my plan progress, modify the upcoming 7 days of workouts” proposed prompt in the chat soon.
Also you got me going on the chat speed I’ll launch a new version tomorrow with an ‘Extended Thinking’ option (similar to Claude or ChatGPT) that will be slower but uses more thinking token and a default version without it that should be much faster. Also includes some other speed optimizations.
thanks great to hear. Basically i’m trying to constrain the plan to something very basic for now. So maybe its not liking that..
i.e. my prompt is sayng keep 90% of my runs below LT1, the rest below LT2. In a very long base building phase. I see its super keen to throw in tempo runs mid-week. But I’ve clearly said - i’m doing one quality track session a week on Monday and that’s it (on my prompt).
The planner seems quite keen to throw in more quality runs than i’d like. So i then adjust on chat.
Please let me know if there’s still anything that’s not good/ideal or broken. It’s hard to test all the possibilities with such a general tool as the chat
