For certain people, a fitness tracker like a Whoop band, FitBit, Apple Watch, Oura Ring, or Garmin smartwatch can provide useful and actionable insights into their daily behaviors. They might feel more dialed in with their marathon training and sleep patterns, or maybe they just like wearing some kind of expensive reminder to live a little healthier. Now, artificial intelligence is showing up in those devices—whether that’s a feature or a bug depends on who you ask.

Whoop’s new feature, Whoop Coach, is powered by OpenAI’s GPT-4 model. This coach, in the form of a chat interface, will attempt to answer the user’s open-ended queries, based on the reams of health data that the Whoop band collects, such as blood oxygen, skin temperature, heart rate, and respiratory rate. Users can now ask “why am I tired?” or “what should I do at the gym today?” and it will spit out an answer. If users cannot stop thinking about the Roman empire and want to work out about it, Whoop Coach will gamely respond with some suggestions.

Whoop seems to anticipate that not all users will be AI-keen, as its product announcement also explains how you can turn the new feature off, if desired. Emily M. Bender, a University of Washington professor, and one of Time Magazine’s most 100 influential people in AI, expressed skepticism, tweeting, “GPT4 is inside this thing. So: are we talking days, hours, or minutes before we see the first reports of physically dangerous advice coming out of it?”

But some people are into it, with positive reviews ranging from “nothing to complain about” to a “dawning of an era of truly smart wearables.” When less than a quarter of American adults are meeting the recommended amount for exercise, maybe a little AI nudge isn’t so bad.

Elsewhere, runners have been asking ChatGPT to create training plans for them. One such plan left a positive first impression on a senior researcher at Polar, calling it a “solid starting point for many runners.” That impression might be the reason that AI takes off in the training world: sometimes it matters to get instructions more than the actual instructions themselves. Perhaps if people feel they’re getting a tailor-made training plan, they’re more likely to adhere to it, even if a hypothetical TrainerBot has three templates that it chucks out to all users.

There is already plenty of interest and investment for AI in the world of cardiovascular health. For one: earlier this month, Cardiosense, a “digital biomarker platform,” started enrollment in a nationwide study to use its AI platform and FDA Breakthrough-designated device to monitor heart failure, a much higher-stakes endeavor than building a couch-to-5k plan.

But there are also reasons to exercise caution. Earlier this year, the National Eating Disorders Association (NEDA) put Tessa—its own AI chatbot that replaced a human-based helpline—on ice after it was found to give dangerous advice around eating disorders.

AI in fitness trackers will seem like a tremendous feature to some people, and more like a pesky bug to others. I fall into the latter camp—I’d already had an on-and-off relationship with Whoop before OpenAI got involved. I’ve found a fitness tracker can quickly become a shackle to a bunch of metrics that occupy a disproportionate amount of my mental real estate. I’m not sure an AI model fixes that.