The most important work I do these days is thinking hard and trying to make good decisions. The people I work with are usually better than I am at implanting those decisions, whether that be fixing farm equipment, talking to customers, or doing accounting. I subscribe to the best artificial intelligence available and have been astounded by its exponential advances to now reach higher levels of intelligence than any humans. It is extraordinarily dangerous for the survival of humanity, but also fascinating to watch and use.
Grok Heavy ($300 / mo) has recently started to really impress me versus Gemini Pro ($30 / mo). Grok Heavy does better research, spends more time on problems, and gives more accurate, and very importantly, more nuanced answers. Gemini tends to get a lot right but then makes arbitrary leaps to bombastic conclusions that are painfully incorrect. I haven’t used ChatGPT as much, but I plan to start. My next programming build (prompt assisted I am sure) will be to put all of these into a meeting room where I can put a single prompt then the AIs will work together to give me solutions.
Here is an example Grok Heavy conversation I had tonight where it parsed six years of fitness data files from Zwift, a bike racing game I do for fun and exercise. I also asked it about the science on caffeine and dark chocolate, which I cut out about a year and a half ago (other than a short period of using again for about a month recently):
## Conversation summary Grok Heavy:
You asked whether **VO2 max is the top marker of longevity and quality of life**, then applied that question to your own life: 42-year-old male, hip osteoarthritis (no running), Zwift racer who lifts, farm owner since Jan 2024, 217 lb since 2022 (210 lb before that).
**On the science:** Cardiorespiratory fitness is one of the strongest *physiological* predictors of mortality in large cohorts (often stronger than many lab values). It is not the only or “top” healthspan marker. Muscle, gait, metabolic health, sleep, and relationships independently matter. Population “elite fitness” claims are often treadmill-METs in clinic patients, not lab VO2 in cyclists.
**On your engine:** FTP and W/kg on Zwift are useful **proxies**, not a lab VO2. Best field stand-in is **5-minute power** (MAP), then 20-minute × 0.95 for FTP. At 98.4 kg, recent racing (~290 W / 5 min, ~260 W FTP) estimates **~43 ml/kg/min** — good / above average for a 42-year-old man, not “high elite.”
**On the full FIT archive (1,357 files, Dec 2019–Sep 2026):**
– Built from ~200 W / 20 min (2019) through a huge 2020–21 hour load, then a **2023 peak**: ~370 W / 5 min (3.76 W/kg), ~322 W / 20 min, ~306 W FTP, Sitko VO2 **~50**, 1-min 582 W, sprints **~1,200 W** (real on your Chrono).
– 2024 volume collapsed when you bought the farm (60 Zwift hours vs 224 in 2021).
– Winter 2025–26 5-min recovered to **~353 W**; threshold did not fully return.
– Now you are ~15–17% down on 5- and 20-min vs 2023, same weight. Ceiling still shows up when you race hard.
**On caffeine and dark chocolate:** Pre-2023 you used coffee and dark chocolate as race tools and paid with mood crashes, irritability toward family, and flutters (also sometimes off caffeine). After cutting them ~1.5 years ago, mood, marriage, kids, and daily output improved. Literature: moderate caffeinated coffee is *associated* with lower mortality and, in a 2026 NHS/HPFS analysis, ~18% lower dementia risk at 2–3 cups; **decaf did not show the dementia signal** in that paper. Caffeine is a proven ~2–5% cycling ergogenic. You are a high-sensitivity phenotype (possible CYP1A2 / ADORA2A pattern — untested). Chocolate adds theobromine (long half-life) plus caffeine.
**On Gemini:** Right call that you should not restart daily caffeine for “longevity.” Too clean on “polyphenols do all the dementia work,” guessed genotypes, “adrenal drain,” and flavanol pills as equivalent to coffee.
—
## Top recommendations
1. **Do not restart daily coffee or dark chocolate** to chase dementia headlines. Your n=1 already showed net harm on mood, family, and (likely) sleep — stronger healthspan effects than an observational HR of 0.82.
2. **Keep the longevity stack you already have:** Zwift + lifting + farm work + stable home life. That combination beats a cup of coffee.
3. **Track fitness with 5-min and 20-min power**, not vibes. Monthly all-out 5 min on the Chrono is enough. Current realistic band: hold **~2.7–2.9 W/kg FTP** and nudge 5-min back toward 320+ if hours allow. 2023’s 3.1 W/kg FTP needs more saddle time than the farm gives.
4. **Train to the constraint.** One 45–70 min ride per week with 2×15 or 3×10 at ~260–280 W is what the files say is missing. Micros/crits already protect 5-min and sprint. Keep lifting for hip, muscle, and the 1,200 W.
5. **Polyphenols without the crash:** tea if you tolerate it; apples/berries; optional Swiss Water decaf as a ritual. Tiny unsweetened cocoa only if mood and heart stay quiet. Skip high-dose flavanol supplements as a default.
6. **Race tools:** nitrates (beet) and carbs first. If you ever test caffeine again, one low dose (≈1.5–3 mg/kg), early, no chocolate, and grade the next 48 hours by family and sleep — not watts. Stop if the crash returns.
7. **Get the flutters documented once** (ECG ± short monitor). They occur off caffeine too; that is a check, not a coffee policy.
**One-sentence close:** You built a ~50 ml/kg rider by 2023, stepped off the stimulant treadmill, and traded peak FTP for a farm and a better household — keep the household, keep racing, add a little threshold work, and treat coffee as optional race chemistry, not a longevity requirement.