Model behaviour: AI coaching after the 10K

After training for my first 10K guided solely by AI not only did I complete it but I also raced in another just two weeks later. Here's how it all unfolded.

Week 1 - Establishing the Routine

Back in January I tasked ChatGPT to figure out how I - a guy who could just about plod his way to 5K - could train up to run the Bournemouth Bay Run 10K event. Giving a goal time of under 1 hour and a schedule when I could train, it generated me a nice mix of easy runs, long runs and speed sessions I could follow day by day.

As a mostly vibes-based runner up until now, I liked how it gave me reasons for why I was doing sessions on certain days. It gave me a good structure to my week with a little bit of flexibility if I needed it.

Despite telling it that I could train any days apart from Tuesday and Thursdays, it still added those as part of the schedule. It became a bit of a running theme. It would remember for a bit until I came back and requested tweaks or follow-ups that required showing a revised schedule. Nothing a quick reminder wouldn't fix, but still a little annoying.

To test the waters a little, I asked it to elaborate more on what "strength training" was. I wasn't sure if it meant hardcore weight training, or if it was more body weight exercises I could do at home.

It turned out to be both. Just giving it a short prompt it generated a set list I could reference day by day. Perfect.

"Could you elaborate more on the strength workouts? What equipment might I need?"

...and now I know what a Bird Dog is. It's not as fun as it sounds, sadly.

The only other tweak I made was to list any speeds in both mins/km and km/h. I prefer mins/km as I find that a bit easier to understand in the real world, but the treadmills at the gym work only in km/h. I asked and then every speed it gave me after that was converted next to it. One less thing to think about, which is ideal!

Week 5 - Intensifying the Workouts

I was firmly in training mode by the fifth week. I'd never committed to any kind of training plan before, but this one I could. I think it was a combination of the easier "base building" block of workouts with just the general novelty of it all.

The first 4 weeks were all about getting the fundamentals in place - locking in the pace, length and cadence to improve upon later. I was ready for week 5 onwards to be harder. It was, for the most part.

Wednesdays were speed workout days. The first 4 weeks were 6 x 400m at a faster pace, with a 90 second recovery in between. It's a push, but it felt a lot easier than other days. Week 5 onwards was the same just with 8 repetitions. It didn't feel much of a step up. I asked it to suggest ways to make it a bit harder.

"I find the Wednesday workout a bit easy. How can I step it up a bit?"

I was expecting it to just recommend that I do more repetitions and it did do that. But it also offered tweaks to speed, distance and opting for more active recovery periods. It even started suggesting things like ladder intervals, hill repeats and threshold runs, which were all bit too hardcore for me so early in that training!

I opted for a mix of greater speed, further distance and more active recovery periods, which it carried over to the rest of the training plan. It even adjusted the tempo and long runs to reflect the faster nature of the speed workout with the encouragement to push harder. I'll take it!

The long runs had became progressively longer. They started at 6km and added 1km each week. There was a bit of confusion around week 5 when it suggested I start at 9km, when from the previous weeks it should have been 10km.

One of the quirks of AI chats is that they seem so eager to prove you right when asking a question. In this case I would have loved a clear answer, but any kind of follow up would change its answer to fit that follow up. Perhaps it's an issue with my prompting, but I would expect someone non-technical would just blindly take the answer they're given. It's a little bit of a concern. In the end I went with 10km because I felt I could do it. No AI help needed.

Week 9 - Final Preparations

This was the beginning of the end of the training. I was supposed to be as good as I was going to get and just needed to get the legs used to the distance - and to be honest it felt that way too!

I asked ChatGPT for a summary of the remaining weeks. Given the previous clarification over the weeks it had initially given me a response as if it was still the fifth week, which confused things a little. But after giving it that day's date it soon corrected itself. It does know what day it is, but must have prioritised the context of the previous session instead.

I'll hold my hands up. I hadn't really committed to the strength training part. I'd mostly focused on the distance. I checked in to see what that meant.

"I haven’t been able to do strength training. Is that bad?"

Instead it suggested I focus on the exercises that assisted running and to take it fairly easy. Given that the strength exercises were to reduce the risk of injury, that seems like a good shout. I wasn't about to try and cram 12 weeks of strength exercises into 3!

But overall the training went well. It was all pretty uneventful heading up to race day. In fact I felt like I could do it quicker, so I asked if that seemed like a good idea. Instead of adjusting the training plan at this late stage, it instead suggested adjusting the pace on race day if I felt up to it. Starting slow and finishing fast was always the plan, but now it was suggesting starting normally and finish faster!

Race day

When it came to the day of the 10K I was as prepared as I could be. So how did I do?

I smashed it. Final time 54:56 seconds. I think my official time was slightly shorter too. Lovely stuff.

Did I enjoy it? I think so. I never thought I'd say I enjoyed running, but there's something about a race day. People cheering you on is a real energiser. Plus I'm a sucker for a kid wanting a high five.

In fact I enjoyed it so much I went straight into another one. The Southampton 10K was just two weeks after this.

"I’ve done the 10K now, but have another one coming up in a couple of weeks. What should I do to keep the fitness up?"

It gave me a plan that was essentially a lighter version of what I had been doing up until this point, with the goal of staying fresh for the next race. If anything it was trying too hard to stick with the original plan. Perhaps the added context was not so helpful here.

But you know what? Smashed that one too. 53:53. Shaved a minute off my Strava time, but the chip time was a good 3 minutes faster still. I felt like I could have gone round again, too. I think the fresh legs served me well.

That time I had AI generate me a playlist to carry me around the course. It took the songs that I like, the pace I'd be running at and an estimated finish time and curated a playlist to carry me around the course. It added some bangers! You can listen on Spotify if you'd like, but I'll go into it in a bit more detail in a future post. Watch this space!

Can AI replace a personal trainer?

The million dollar question.

I had a blast with this challenge. It helped me learn more about AI, about running and generally how disruptive AI can be to pretty much any activity.

I think for 90% of people it would get them 90% of the way there. That's a massive impact. Instead of just guessing at it or downloading some generic training plan from the internet, everyday people like me can get tailored training plans for free and instantaneously. It's a little eager to please, but I'll take that over some screaming PT that's really not getting my energy.

However, it's not going to suit everyone. Nor is it it going to get people across the line. That's where a personal trainer is going to excel. They will know their client and where their strengths (and weaknesses) lie. If anything is going to get the best out of someone, it's a person who has trained for years just to do just that.

When I started this experiment it's safe to say I was skeptical. But after sticking with it for 3 months, I began to understand it more. I would get more out of this process if I put more in. If I learned how AI could help, I could choose my questions and structure my prompts to get the best results for me. It's solidified my thoughts that AI is a tool and not just a replacement for everything.

If you've ever considered giving running a try, get an AI to help you out. Even if running's not your thing, it's shown me that it could encourage and help you improve on pretty much any activity as long as you give it the chance to. Sure, it could write your novel or paint that picture, but that's not going to improve you. That one's on you. But now, you don't have to be alone in that. That seems to me like a good thing.