Nicholas Thompson, magazine editor turned tech exec, has quietly revolutionized personal fitness with AI. The guy took his marathon times from a respectable 2:40 to a jaw-dropping 2:29 while in his 40s—when most runners are shopping for knee braces and reminiscing about “the good old days.” How? By creating a custom GPT coach loaded with his personal training data.
While others his age shop for knee braces, Thompson used AI to smash marathon records in his 40s.
This isn’t your basic “drink more water” fitness app. Thompson’s AI analyzes his unique patterns and spits out real-time coaching that mirrors what elite Nike trainers provided him earlier. It’s personalized to an almost creepy degree, recognizing exactly what works for his specific body. The system doesn’t just count steps; it thinks like a coach with years of Thompson-specific knowledge. Much like AI virtual assistants now manage complex tasks beyond simple commands, Thompson’s custom coach handles intricate training adjustments with remarkable precision.
The timing couldn’t be more relevant. We’re standing at the precipice of the next AI wave, with trillion-parameter models about to hit the scene. GPT-5 looms on the horizon, following the soon-to-be-released GPT-4.5 “Orion”—the last major non-chain-of-thought model before things get really wild. Though OpenAI has officially stated that GPT-5 is not currently training, numerous reports suggest preparations are well underway. Scientists can now predict AI intelligence before training even begins. That’s like knowing a kid’s SAT score at birth.
The implications stretch far beyond marathons. By March 2025, AI will reportedly approach 94% of AGI benchmarks. Brain-machine interfaces from Neuralink and Synchron are already being demonstrated. Companies are scrambling to integrate AI agents into workflows, though most businesses still haven’t figured out how to use them effectively. No surprise there.
What’s fascinating is the accelerator effect. Thompson’s personal experiment shows how AI can maintain and improve performance once humans hit their natural limits. It’s a microcosm of what’s happening in the broader AI landscape, where capabilities are expected to “steamroll” slower development efforts. New projections indicate GPT-5 will be trained on an astounding 50 trillion tokens of synthetic data.
The race is on—both literally for Thompson and figuratively for AI development. The question isn’t whether AI will transform industries, but which humans will be smart enough to harness it first.
References
- https://www.techbuzz.ai/articles/atlantic-ceo-built-custom-gpt-coach-to-smash-marathon-records
- https://lifearchitect.ai/gpt-4-5/
- https://lifearchitect.ai/gpt-5/
- https://lifearchitect.substack.com/p/the-memo-12apr2025
- https://stratechery.com/2025/ais-uneven-arrival/
- https://mastersofscale.com/episode/big-tech-trends-shaping-2025/
