I'll just never feel the same about "insights, advice and support" coming from a LLM vs. a real person. I know that perhaps in a blind A/B testing via text channels I probably won't be able to tell the difference, at least not right off the bat, but there's just something profoundly wrong with humanizing this. Personally I'd very much like an "AI training framework" (tried many but no fit yet), but I don't need a digital cheerleader.

RE: calorie counting, I used to work on a calorie counting app that worked on delivery meals (back in China - nobody cooks there). It was way back then, so we talked to the chefs and had recipes w/USDA-like DBs to estimate calories. The assumption was that you measure your intake from the app and your weight, if it doesn't go up/down as expected, you just adjust a coefficient every N period, i.e. not losing weight? ok let's set all calorie estimates at 1.1x for you. Rinse and repeat until you're losing weight.