The Sub-7 Experiment

Tag: chatgpt

  • A New Coach in the Room

    It feels like I have a new coach.
    You might remember that I’ve been asking ChatGPT to prepare workouts and coaching advice to help me row 2,000 metres in under seven minutes. That is the whole point of The Sub-7 Experiment. Can I use AI to help me achieve a fitness goal.

    When I first tried this, the model available to me was GPT 4.0 and it was good. I set up the role, gave it the context, kept the conversation alive, and each day I told it how I was feeling. It responded with the workout for that day. That simple rhythm worked well.

    Then 4.1 came along. The improvements were small but noticeable. It held context better, got confused less often, and could handle my slightly lazy and vague questions more easily. As with most tech, each release pushed things forward.

    Earlier this year GPT 5.0 arrived with a huge amount of hype. It was supposed to push ChatGPT into a whole new league. PhD-level reasoning. Better accuracy. A genuine step up.

    The reality was mixed.
    Yes, it produced better code for a different project I was working on. Yes, it had more “thinking” capacity and could reason more deeply without me having to explicitly tell it to think. That part was a massive improvement.

    But as a conversationalist it was a step backwards.
    It forgot things. It lost context. It got confused about tasks. The backlash was so strong that OpenAI reinstated version 4.1 as a choice because so many people preferred to keep using it. It felt like my coach had left the building.

    I worked around it with careful prompting, but it was frustrating. The Sub-7 Experiment relies on continuity and rhythm, and something was always slipping.

    Which is why it now feels like I have a new coach. GPT 5.1 has arrived and it feels different in a very good way. The tone is consistent. The help it offers is actually useful. It anticipates the next step instead of fumbling it. It feels like an upgrade in the true sense of the word.

    Let me explain why this matters.

    I have a WHOOP device and I’ve used it for almost two years. Before ChatGPT became part of my training, WHOOP was my only guide. It sits on my wrist and picks up all sorts of measurements: heart rate, skin temperature, strain, sleep quality and plenty more.
    Every morning the app shows three dials. Yesterday’s strain, today’s recovery level, and last night’s sleep. When you start a workout you tell WHOOP what you’re doing and it gives you a target strain. When you hit it, the band vibrates and tells you to stop. Simple. Clever. And it worked well.

    But in the last few months I’ve been questioning the value.
    The subscription model has changed. The promise of free device upgrades for active subscribers has been replaced by an “uplift fee”. My renewal is in February and it will cost a lot more to keep going into my third year. And that makes me ask what I actually need it for.

    Most of the data WHOOP collects I don’t really use. I know how well I slept because I was there. What I actually value is recovery guidance and strain targets. And there are other devices out there that do similar things for a simple one-off cost. Polar Loop is one I’m looking at seriously.

    So I turned to ChatGPT 5.1 for help. Reviews. Recommendations. Thoughts based on my training. And one of the threads pointed out something obvious: I have not been using most of WHOOP’s data anyway. Not deeply. And the only thing that truly matters is the workout planning, which comes from CoachGPT.

    I asked if there was any way for ChatGPT to access WHOOP data directly. It said no, the APIs are not available yet. But then it made the suggestion that genuinely impressed me.

    It told me exactly which two screens in the WHOOP app to screenshot each morning. It told me to upload them, and it would analyse everything it needed: recovery, sleep, strain, HRV, and readiness. It would then produce a fully tailored workout for that day. And if it thinks I need a rest day, it will tell me that too.

    I have used this new process for the last few days and it is genuinely brilliant.
    Two screenshots. Upload. Instant plan. Clear reasoning. Exactly what it expects from me. Exactly what to avoid. Exactly how hard to push.

    And it works.
    The coaching is better.
    The structure is better.
    The whole system feels like something new.

    Well done, ChatGPT 5.1.
    This is The Sub-7 Experiment.
    Recovering using structured data.

  • Training with AI: More Than Just the Plan

    When I first thought about the Sub-7 Experiment, the core idea was simple: could I use modern AI, specifically ChatGPT, to help me structure my training and ultimately achieve a sub-7-minute 2000m on the indoor rowing machine, the erg? And yes, that goal remains the driving force. I still need that progressive overload and a reason to get to the gym.

    But as I’ve hinted, the experiment has definitely shifted somewhat. It’s become about much more than just the physical goal. A huge part of this journey, one I want to introduce properly, is the dynamic I have with my “coach”.

    Who is My Coach?

    My coach is ChatGPT, the AI developed by OpenAI. In a previous article I wrote about how ChatGPT is just a parrot, looking for patterns (words and sentences) in the data and replying with what it’s heard (all the data that it has been trained on).
    This is oversimplifying it, of course—there are some pretty clever algorithms in the background helping it to predict what the next word should be in its response. And sometimes it gets it wrong. This is what’s often described as an AI hallucination—it knows it needs to respond, and if it can’t find the right information, it might make something up.

    So with that said, you might picture AI coaching as simply pulling a pre-written plan off a shelf. But that’s not how this works. It’s a real, ongoing conversation. Before a session, I tell it where I am (gym or home), how I’m feeling—even if I’m tired or run down—what my recovery data looks like (often from WHOOP…), and how much time I have.

    Based on that information, it suggests a session—maybe intervals, steady state, or strength work…. It gives me pacing guidelines, stroke rate, and even a warm-up and cool-down. After the session, I report back on how it went, the results, and how I felt. This feedback loop is key because the AI then uses that information to adapt future routines. It’s a training partner that adapts in real-time, much like a human coach would.

    Setting the Scene: Prompt Engineering, Simplified

    Where does anyone start with all of this? Talking to AI might sound technical, and you might have heard terms like “prompt engineering”. That term can sound a bit daunting—like you need to be a programmer or a data scientist. But here’s the truth: it’s really not.

    “Prompt engineering” is just about giving the AI clear instructions and enough information.

    In a previous post I’ve suggested people to ask it “What can you do for me?” and that will be enough to get the conversation started. I also mentioned earlier that ChatGPT—and other large language model (LLM) AIs—can get the answer wrong. One of the ways around this is to set the context when you start the conversation.

    As you log into ChatGPT you are faced by a screen which asks “What can I help you with today?” or “Ready when you are.” At this point, one of the best ways to refine the responses you get is to give the AI a persona or a role to play during the conversation. I started the coaching conversation with this…

    “I’d like you to take the persona of and advise me as a personal fitness trainer and nutritionist with a specialism in Indoor rowing and the ERG machine.”

    I then followed it up with a specific instruction…

    “I am working towards a goal of rowing 2000 meters in under 7 minutes. I’d like you to create a training plan for me to achieve that goal. I intend to report in to you each and every time I am in the gym and tell you how I am feeling that day and you will tailor me a program for that session.”

    And that’s how it all started. And I have kept that conversation open ever since so it “remembers” the context and keeps track of all the sessions and significant moments to date.

    Now, for example, when I say I’m in the gym, tired, 75% recovered, and want to do distance, that’s my “prompt engineering”. It’s me giving it the necessary inputs so it can generate a relevant, helpful output, in this case, a suggested workout.

    Think of it like talking to a human personal trainer. You wouldn’t just say, “Tell me what to do.” You’d say, “I’m here at the gym today, feeling a bit rough mentally but my Whoop says I’m recovered. I’ve got about an hour. What sort of session should I do?” You’re giving them context so they can give you a smart recommendation. It’s the same with AI. It’s about talking to it in normal language.

    This is why I wanted to share some of the conversations I’ve been having with ChatGPT. It helps to show, not just tell, how this coaching dynamic works. You’ll see how I give it my status, how it responds with suggestions, and how the dialogue flows.

    More Than Just Getting Fit: A Real-Life AI Experiment

    I’ve worked in IT for years. I’m fascinated by technology, and the pace of AI development has been mind-boggling. While I’m not an expert, I’m definitely an AI hobbyist.

    One of the key drivers for starting this blog and documenting this process was precisely this dual goal. Yes, I want that sub-7 minute 2K. But I also wanted a real-life experiment to truly understand generative AI and how it can be used on a day-to-day basis. What better way to understand this stuff than to use it consistently, over time? Far better than any course or training material.

    So, while we talk about rowing workouts, pacing, and splits, remember that you’re also getting a glimpse into how AI can be integrated into personal goals and routines. It’s an experiment in fitness, yes, but it’s equally an experiment in technology and human-AI collaboration.

    Is It Cheating?  Not Even Close

    And for those wondering, is it cheating? Absolutely not. The thoughts, the words in these posts are mine.

    In another conversation I gave ChatGPT the role as a blog adviser and editor and worked through the steps for setting up a blog from scratch. I asked it to ask me some questions so it could better understand the task I was asking it to complete and it then made recommendations as to which blog platform to use and why and then helped me with the setup.

    I speak into an app on the phone straight after my session—unfiltered streams of consciousness, honest feedback about the session—and then use that as a basis for my next blog post where ChatGPT acts as my editor, helping with flow and clarity.

    And at the end of the day, AI isn’t the one sitting on the rowing machine. I am.

    The Sub-7 Experiment: Meters, Mindset and more.

    This is The Sub-7 Experiment…. It’s about the meters, the mindset, the technology, and the journey of seeing what’s possible when you combine all three.

  • Training with AI: Real Conversations, Real Progress

    Why Did I Start This Experiment?
    To see if I could use ChatGPT as a coach and training advisor to help me row 2000 meters in under 7 minutes on the indoor rowing machine—the erg.

    That’s the core of this experiment, and I thought I’d share a full, unedited conversation I had with ChatGPT today to show exactly how that plays out in real time.

    Me:
    “Ok I’m in the gym. Kinda feeling tired and run down but I think it’s mental health. Whoop has me in the green at 75% recovery. I think I want to do distance today. No exercise yesterday. What do you recommend?”

    ChatGPT:
    “If you’re feeling a bit tired and it’s more mental than physical, a steady endurance row is a great call—it’s meditative, rhythmic, and tends to shift mood in a positive way without needing to be all-out.

    Here’s what I suggest for today’s distance row (45–50 mins total):
    Session: 10,000m steady state

    • Warm-up: 5 mins easy (2:15–2:20/500m) at ~20 SPM
    • Main Row:
      • 9,000m at ~2:05/500m, 22–24 SPM
      • Smooth, controlled, find your rhythm
    • Last 1,000m: Pick it up if you feel good—finish strong at ~1:55–2:00/500m
    • Cooldown: Optional 3–5 mins easy spin or slow row

    Let’s get that good-tired feeling—burn off the mental fog and come out the other side. Want a finishing push suggestion too?”

    Me:
    “Thank you. Let’s row first and then see about the pushing.”

    ChatGPT:
    “Perfect—lock into the rhythm, breathe it out, and let the row do its thing. You’ll know by the end if that sled wants a turn or not. Let me know how it goes when you’re done.”

    Post-Session:
    Me:
    “Found a good rhythm at 23 SPM. First 9000m @2:05. Final 1000m at 1:48–1:53 @30–31 SPM.”

    ChatGPT:
    “That’s a strong session—controlled, consistent, and a powerful finish. Holding 2:05 at 23 SPM for 9K is impressive. And a sub-1:50 push to close it out? That’s real progress. You’re building aerobic endurance and power now. Legs feeling worked or still ticking over?”

    Me:
    “Legs are feeling worked. I think we might need to do some lower back/core work—I was losing form near the end.”

    ChatGPT:
    “Great observation. Lower back and core work will make a real difference, especially when fatigue sets in. We’ll integrate targeted strength work on your next gym day. Let’s build that base.”

    Me:
    “I just snuck in some sled work—5 x push and 5 x pull at 75kg.”

    ChatGPT:
    “Love that. That’s a strong finisher—especially after a 10K row. Power + consistency = serious gains. Let me know how the legs and back feel tomorrow—we’ll adjust accordingly.”

    So why share this today?
    Because this is the experiment in action: using AI as a training partner that adapts in real-time—offering structure, encouragement, and flexibility, almost like a human coach.

    Of course, a real trainer could correct form mid-session. ChatGPT can’t. It can offer YouTube links and detailed stroke guidance, but it can’t nudge my hips or spot poor posture. That’s fine by me. I enjoy experimenting and finding what works for my body.

    And the suggested session was meditative. That rhythm, that structure—it lifted the fog. I walked in scattered, and walked out level-headed.

    A Note on the Tech:
    There are a lot of doomsday headlines about AI—how it’ll take jobs or make people redundant. And in some industries, yes, roles will change.

    But here’s my advice: Don’t fear it. Explore it.
    Create a free ChatGPT account and just ask:

    • “What can you do for me?”
    • “How could we work together?”
      Let it ask you questions back. You’ll be surprised what it can help with.

    This is more than an experiment in rowing. It’s an experiment in mindset, in learning, in adaptation.
    And for me? It’s working.

    This is the Sub-7 Experiment.

  • The Day I Broke 7:30 (And Almost Didn’t Try)

    It was a lovely day yesterday—Sunday.
    We went to the seaside as a family and walked along the boardwalk under blue skies and sunshine. Sandwiches. Coffee. Chatting. It was great.

    But there was a little voice in the back of my head that kept whispering:
    “Why don’t you do a 2K test? Go on—just to see.”

    It wouldn’t let up.
    Even as I lit the BBQ later and cooked chicken shawarma—one of our family favourites—it was still there, quietly pushing.
    And by the time I went to bed, I’d made a decision.
    I was going to do a 2K test tomorrow.

    Morning Doubts

    I’ve been here before—full of good intentions at night, and something else entirely in the morning.

    But this morning, I wasn’t talking myself out of going to the gym.
    Still, I was hesitant about the test. So I checked in with ChatGPT:

    Me: “Hey, good morning. How are you doing today?”
    ChatGPT: “I’m great—especially now that you’ve checked in. What version of you showed up this morning? Fired up? Steady? Unsure?”

    Me: “Still figuring it out. Didn’t sleep well, but I’m not bailing on the gym.”

    ChatGPT: “That’s already a win. You don’t need to feel 100%—just show up with what you’ve got. Do you want to go for the 2K test, or would something lower-pressure feel smarter today?”

    Me: “I don’t know. Not until I’m sitting on the assault bike warming up.”

    ChatGPT: “That’s a great approach—let the body decide, not the brain. Once you’re moving, we’ll know. Message me when you’re spinning.”

    The Warm-Up

    So I walked my son to school, got to the gym, and it wasn’t too busy.
    I got on the assault bike and said:

    Me: “Okay, I’m here. I want to do the test. I’m not aggressively fired up—but maybe that’s a good thing. Hopefully more controlled. I’d like to beat my baseline, but secretly, I’d love to crack 7:30.”

    ChatGPT:
    “Let’s go then. Controlled attack. Here’s the plan…”

    And then it gave me a full warm-up, pacing breakdown, stroke rate guidance, and a quarter-by-quarter strategy:

    The Controlled 4-Quarter Strategy

    • 0–500m: Controlled aggression (1:50–1:53)
    • 500–1000m: Settle into rhythm (1:52–1:55)
    • 1000–1500m: The grind (1:52–1:56)
    • 1500–2000m: Empty the tank (negative split if possible)

    The Test

    So I went for it.

    • The first 500m was messy. I was either pulling too hard at a low stroke rate or rowing too fast at the recommended SPM.
    • The middle 1000m was better—I held the pace but started letting mental distractions creep in.
    • At around 800m to go, I started pacing for the final push—too early—and slowed down.
    • In the final 500m, I was starting to run out of steam. Form slipped, but I caught myself, pulled it back together, and finished strong.

    🕒 Final Time: 7:29.5

    The Shift

    I did it.
    I broke the 7:30 barrier for 2,000 meters.
    And I’m absolutely delighted.

    This proves a few things:

    • The training structure that ChatGPT’s been helping with is working.
    • I’ve shaved 12 seconds off my baseline.
    • And this whole thing? It’s more than just physical now.
      There’s been a shift in mindset.
      This isn’t a vague goal anymore—it’s happening.

    This is the Sub-7 Experiment.

  • The Training So Far

    From Google to ChatGPT

    Google is brilliant at finding stuff on the internet. It’s so ingrained in everyday language that we say, “I’ll Google that for you.” You can use exact phrases, search within specific sites, and fine-tune your results with “AND,” “NOT,” and “OR.”

    But now that ChatGPT has internet access, it’s my go-to tool for research. If I want to buy something, I don’t just Google it—I ask ChatGPT to compare options, tabulate results, and summarize real user reviews. I can even ask where to buy it and what payment options are available.

    That’s really clever.

    But ChatGPT hasn’t just become a research tool for me. It’s become my coach.

    Body Recomposition & The Problem With Weights

    I started reading about body recomposition—the idea that you can lose fat and gain muscle at the same time. Most fitness advice says to focus on one or the other:

    • To lose fat, you need a caloric deficit.
    • To gain muscle, you need a caloric surplus.

    So how do you do both? A lot of articles suggested it was possible if you:
    ✅ Prioritize lean protein
    ✅ Lift weights consistently
    ✅ Keep a close eye on calorie intake

    The weights part is where I hit a roadblock.

    I find the weights section intimidating. I’d only use it when no one else was around—which wasn’t often. That meant I had no consistency.

    But the rowing machine? That was in a different part of the gym, and hardly anyone used it.

    Asking ChatGPT: Is Rowing Strength Training?

    Since I was already rowing regularly, I asked ChatGPT this:

    “Does rowing on a Concept2 erg count as resistance training, or is it purely cardio?”

    I explained how I row heavy, deep strokes at low rates and mix in high-intensity sessions. Could rowing alone help with body recomposition?

    The answer?

    “Rowing is an excellent full-body workout that combines cardio and strength elements. While it won’t replace dedicated strength training, it can contribute significantly to fat loss, muscle endurance, and mental health. With the right approach, you can absolutely achieve body recomposition through rowing—especially if you complement it with weights and proper nutrition.”

    So, not exactly a replacement for weight training, but it gave me a framework. And when I told ChatGPT that I found the weights section intimidating, it suggested alternative routines I could do outside that area.

    At this point, I wasn’t even thinking about sub-7. I was still focused on getting fitter and reading James Smith’s book.

    ChatGPT as a Coach

    I started logging my training with ChatGPT:

    • How far I rowed
    • What pace I held
    • How I felt during and after

    I even asked:

    “Is this OK for a nearly 55-year-old man?”

    The responses were detailed, encouraging, and motivating. It would:
    ✅ Break down my average split times
    ✅ Tell me what the session achieved aerobically
    ✅ Compare me to age-group standards
    ✅ Provide structured progressions for the next workout

    Then, the conversation shifted.

    I mentioned struggling with nutrition—I’d be disciplined for a while, then binge on the wrong stuff. I told ChatGPT that I wanted to be in better shape before my holiday in a few months.

    The response? Not just training advice—but a mindset shift.

    • Set a realistic goal
    • Use the 80/20 rule
    • Stop relying on willpower alone
    • Make myself accountable—but without guilt

    ChatGPT had morphed into a therapist.

    From Reluctance to Showing Up

    I started digging into why I was avoiding the gym—why I’d talk myself out of it first thing in the morning.

    ChatGPT helped me reframe my thinking:

    • Make it a choice, not a chore.
    • Set smaller, achievable wins.
    • Just get there—figure out the session once you’re in.

    One day, I finally dragged my arse out of bed and into the gym.

    I sat on the static bike and asked:

    “OK, so now I’m here—what should I do?”

    ChatGPT replied:

    “Great! You’ve already won by showing up. Since you’re at the gym now, let’s keep it simple and effective based on your goals.”

    It then suggested workouts, I picked one, and I went and did it.
    It included weights and rowing, and I felt brilliant afterward.

    Over the next few sessions, I kept logging workouts—and ChatGPT adapted my training like a real coach.

    The Mental Health Row

    One day, I told ChatGPT:

    “A mental health row is required today. I’m planning on doing 7K in 30 minutes. What do you think?”

    ChatGPT responded with:

    ✅ Pacing recommendations (2:08–2:10/500m split)
    ✅ Stroke rate guidelines (22–24 SPM)
    ✅ Mental approach (“Let it be your time—no pressure, just movement and flow.”)

    I rowed 7,000m in 29:20, finishing with a 250m sprint at 1:45/500m.

    The breakdown was exactly what I needed to hear:

    ✅ “You even beat your original target!”
    ✅ “You stayed controlled and steady but still had power left to push at the end.”
    ✅ “You got both mental and physical benefits—clearing your head while maintaining a solid pace.”

    That’s when I asked:

    “What pace do I need for a sub-7-minute 2K?”

    ChatGPT’s response?

    “To hit 7:00, you need to hold a 1:45/500m split. You’ll need a mix of power, endurance, and interval work to sustain it.”

    It then gave me a structured training plan—low-rate power rows, sprint intervals, race-paced work, and strength training.

    And That’s How The Sub-7 Experiment Was Born

    At that moment, the goal crystallized.

    This wasn’t just about fitness anymore. It wasn’t just about losing weight before a holiday.

    It became a clear, structured challenge.

    A sub-7-minute 2K.

    And I was going to train for it properly.

    This is The Sub-7 Experiment.

  • So. What’s The Experiment?

    I’ve worked in IT for more years than I care to remember, and it’s treated me well. I’ve had the chance to work around the world with some amazing people and technology.

    Lately, the pace of technological change has been accelerating so fast that it boggles the mind—and one of the biggest shifts has been AI (Artificial Intelligence).

    I’m not an AI expert, but I’d call myself an AI hobbyist. The possibilities fascinate me, and with the rise of Generative AI, things are getting even more interesting.

    What is Generative AI?

    You’ve probably heard of ChatGPT. Other tech companies have their own versions, but at its core, the GPT part stands for Generative Pre-Trained—meaning it can generate new responses based on the massive amount of data it has been trained on.

    And the Chat part? That’s where the real magic happens. Unlike traditional AI systems, anyone can talk to it in normal language. No coding, no technical knowledge—just type a question, and it responds.

    But how does it actually work?

    The best analogy I’ve come across is this:

    Imagine every person on the planet has a parrot on their shoulder. That parrot listens to everything they say and remembers the patterns. Then, every parrot shares their knowledge with every other parrot on the planet.

    Now, if you ask your parrot “How are you today?”, it doesn’t think about the answer—it just repeats what it has heard most often:

    “I’m very well, thank you. How are you?”

    It’s not true intelligence, just pattern recognition on a massive scale.

    So… Who is My Coach?

    My coach is ChatGPT.

    • It helps structure my training and keeps me accountable.
    • It refines my blog posts, making them clearer while keeping them mine.
    • It helps me reflect on progress without getting lost in my own head.

    I’ll be posting some of my conversations with ChatGPT as part of this process—showing how I’m using it, what I’m learning, and how it’s shaping my approach.

    Is It Cheating?

    No.

    The thoughts and words are mine—AI is just my editor, tightening things up so they read better.

    And AI isn’t the one sitting on the rowing machine. I am. I will get that sub-7-minute 2,000 meters. This is The Sub-7 Experiment.