AI Endurance adds chatGPT-powered features

AI Endurance has been providing true AI/ML services for triathletes, cyclists, and runners for a couple of years. It’s definitely ahead of the curve when it comes to state-of-the-art training prescriptions. But even in recent weeks, the platform has made further steps forward by incorporating ChatGPT.

“ChatGPT’s impressive capabilities are already enabling features that would have been hard to imagine just a few months ago. We are just at the beginning of AI revolutionizing the triathlon, running and cycling world.” [Markus, aiendurance’s CEO]

Over the last couple of years on this blog, I’ve been approached by a couple of training apps claiming to be AI models. They just aren’t. It’s one thing to use the abbreviation ‘AI’ in your marketing and it’s another thing to actually use it on your platform.

How AIEndurance Uses Machine Learning and Artificial Intelligence

AIEndurance uses machine learning (ML) and artificial intelligence (AI) to personalize training plans for athletes in 5 key areas

• Machine Learning (kinda): The recovery/readiness status is modelled using a flavour of the widely-used Banister model, which is based on load, heart rate variability, muscle soreness, and your report. This produces a percentage recovery value.s (More: https://aiendurance.com/blog/your-heart-rate-variability-recovery-model)
• ChatGPT is used to gauge sentiment by subjective inputs of your motivation, mental state, and so on. Like this…

• AI Endurance automatically detects your performances from links to platforms like Garmin Connect or Strava, or simply from FIT files. It then makes iterative predictions on this data and produces a training plan along with predictions of your future performances.
• Rules are applied to the subjective inputs and the inferences made from them, and to the Banister models to offer you adaptations to the plan. For example, “Your readiness is 10%. Do you want to skip today’s threshold run?” Depending on your readiness state, varying degrees of severity are applied to the adaptations offered.
• ChatGPT is used to gauge sentiment by subjective inputs of your motivation, mental state, and so on. It responds to several inputs to the model and determines plan adaptations, including a change to today’s recommended workout, coupled with an explanation of the changes.

Simplistically, machine learning is used for numerical calculations and chatGPT’s AI is used for interactions based on language. These two areas of the model overlap and interact.

One aspect I find interesting is trust. Some of us will unquestionably trust computers’ recommendations, whereas others will be hugely sceptical. AI Endurance shows you its own model’s accuracy over time at predicting some of your performance characteristics such as FTP and rFTP.