Whoop provides Wearable with OpenAI Coach for the first time
There is no shortage of wearables that record the fitness and health status of the user. The US company Whoop also offers a corresponding bracelet, relies on data collection around the clock and sees the tracker above all as part of an overall concept (with a subscription): The user keeps a daily logbook in which, among other things, he records whether he took coffee or medication, dared to look at his cell phone in the evening or was sexually active.
From the measured values and the user’s information, Whoop not only creates evaluations on sleep, recovery and stress on the associated app, but also gives concrete instructions to improve the body’s performance.
Now Whoop is rolling out an important update: The app now has a virtual AI coach who gives detailed and individual answers to health and fitness questions in over 50 different (freely selectable) languages. The basis is GPT-4, the current version of OpenAI’s generative artificial intelligence (AI). c’t was already able to try out the new AI function in the beta version.
First sessions with the AI coach
The integration of the AI function in the app is unspectacular: an additional field on the start screen takes you to a chat window in which you can type in your questions if you have an internet connection and receive answers shortly afterwards. In general, this worked well, but with more complex questions, the virtual coach occasionally faltered at first and even stopped completely. However, a clear improvement has already been noticeable here in the past few days.
First of all, the virtual coach is a database of information about fitness and health as well as a virtual support agent for Whoop that answers questions like “What does heart rate variability mean?” or “My WHOOP has stopped loading – can you help me?” answered – often with links to further articles that can be read directly in the app.
What is more interesting, however, is that you can also ask questions that evaluate your own data. For example, if you ask how you can improve your sleep quality, the AI coach not only gives general advice – such as scheduling time for sleep, darkening the room or reducing caffeine intake – but also looks in your personal logbook to see whether you have been sleeping in the last few days Time spent using a cell phone in bed or taking certain preparations (such as magnesium) in order to generate suggestions.
If you are not sure whether your performance is within the “normal range”, you can ask the coach about it. When asked “How does my sleep compare to other people my age?” For example, I learned that I have a lower need for sleep than the average man my age, but on the other hand, I sleep longer.
According to Whoop, I’m doing a great job at sleeping! But what if I still feel tired? I asked the AI coach.
According to Whoop, the AI coach evaluates thousands of personal data points from its users to find out why they feel the way they do – and how exactly this influences their own performance. When asked by c’t, the company explained that protecting this data was a top priority and that it would ensure that OpenAI did not use the data for other purposes. Whoop claims to be GDPR compliant, but this is not necessarily the case with OpenAI.
According to Whoop, the AI coach can ultimately provide tailored training plans and recommendations, for example for an upcoming 5-kilometer run that you want to complete in 24 minutes in the coming month. However, I did not receive the desired response to these inquiries and was simply referred to member service. According to Whoop, this issue is known and will be resolved with the current firmware update for the band. However, this is not yet available for my test device.
With the current firmware, the question about training plans only generates a message to contact member service.
The reference to member service is an emergency brake that Whoop pulls when the AI reaches its limits – apparently to prevent the coaching function (like ChatGPT in some cases) from slipping and starting to hallucinate. As expected, the virtual coach was also trained primarily with data about fitness and health as well as the Whoop concept. Consequently, he also answers some general questions such as “What is the capital of South Africa” or “What is the size of the earth?”, but immediately brings the conversation back to its core area.
Whoop does not see itself as a pure wearables provider, but rather sells complete health and fitness coaching – at prices starting at 18.50 euros per month (with a 2-year subscription). Instead of general health tips, you can expect individual coaching. In this respect, the manufacturer’s current step is understandable: Since not every customer can be provided with a human trainer, the AI coach is currently the next best option.
From the customer’s perspective, the AI function is first and foremost user-friendly: you don’t have to click through any glossaries or lists to find out what a term means or how to increase its performance.
I’m currently a bit conflicted about the answers: they are undoubtedly tailored to me and the information I provided in the logbook, but so far the recommendations often remain quite general. However, as described, I have not yet been able to access the training plans. The AI coach also needs a lot of basic data, which in my case Whoop hasn’t collected all of it yet.
A detailed test of Whoop’s AI coach will be published in an upcoming c’t. (nij)
To the home page
#Whoop #Wearable #OpenAI #Coach #time