Well, guess how the comfy OK Google or hey siri works, hearing you while you say it all across the room. Or that noise cancelation for your calls. Admittably, the way he’s put it sounds really tinfoil hat weird, but he’s got a point there.
Any current mobile phone is so very crammed with sensors of any kind, which do make a lot of features possible/usable/comfortable and the same sensors may be used to track a good lot of your behavior, if used for malicious purposes. And we know that for a fact with targeted ads, where several people I’ve talked to noticed the same, where that even talking about a topic may be enough for ads to be show up. Check adssettings.google.com for example, it’s actually scary what Google “assumes” about you, and even scarier how on point those assumptions are. A lot of this information is sourced from your devices sensors, and the argument of “there’s just not that much computing power to process this data” is simply not valid anymore.