Show HN: Workout.cool – Open-source fitness coaching platform

I count calories and use chatGPT to do so by taking pics of my meals / what I eat.

I'm looking forward to Meta adding this feature to their Ray Ban smartglasses so the glasses automagically count my calories each time it sees food on my plate or going into my mouth. A feature they possibly should make optional, but for me who has prescription Metas it would be a big time saver (try to eat 1500 to 2000 calories a day and burn 250 to 500 in exercise). I think the knowledge of how many calories you consume done automagically would prompt 1/4 of people thinking and or on Ozempic to not do it.

You know counting calories by taking a picture literally doesn't work right? In fact, I'd argue it's one of the best examples of when an LLM confidently feeds you misinformation. An obvious example, a chicken breast fried in oil and a chicken breast air fried look identical, yet the oil fried one would have 3-4x the calories at least. Don't take my work for it though - this was just on HN the other day: https://news.ycombinator.com/item?id=44220135

What happens if you give it that context though? Or prompt it to search the web for comparable meals?

Well i eat out almost always at healthy-ish chains (Cava, Panera, etc) so GPT gets the calories directly from each chains website. As for food prepared in regular restaurants I use it the same and its not precise but close same with home cooked meals per my testing the data.

Overall I am a bit obsessed but not that obsessed to the point it needs to be exact on-point precise.. just give me an idea of where my calorie count stands for anytime of the day. Just be way better if it was done auto-magically and Im betting this will be a good future use that gets people excited for smart glasses one of many upcoming innovations with them.

I agree in principle, but I'll say that the number of calories are likely not 3-4x. The 3-4x number is for deep frying chicken. You usually don't do that if you aren't also breading the chicken.

A pan-fried chicken with a little oil in the pan to avoid sticking/make better thermal contact will add calories but not 3-4x more. You're likely using about 1 tbsp of oil which is around 100 kcal. 100g of chicken has around 160 kcal. Even assuming all the oil ends up on the chicken (it isn't) that's ~2x the calories at most.

Perspective wise, though, it'll by the white rice or mashed potatos that are more problematic in terms of calories. Both have a load of calories and can't be eyeballed by camera. It's all about the weight for those. And if you threw in butter/oil, even harder to know what the actual calories are.

At that point I just have to ask what the AI is actually doing for you?

I can just hit a search engine and say "number of calories in X" and get a precise answer with 2 seconds of calculator math.

If I have to take a picture, send it to ai, but then amend it with "This is air fried chicken, it weighs x, it's a breast cut. I didn't add salt."

Why do all that when a single search will give me the answer I want without the picture upload or context?

Not if you eat out at well known healthy chains they already have their nutrition details on their sites that AI pulls from.

I agree but with smart glasses doing it for you i feel it's a whole different experience and utility.

Even a little bit of variation in the amount of calories consumed vs detected by AI will be the difference between gaining or losing weight. Also, since fats are the most dense calorie wise (9 per gram), and also hardest to detect by AI as fats are usually transparent (oil), it's even harder to get a remotely accurate measurement of calories.

I think most important thing is calories consumed not fats and sugars, etc as it's simple the less you consume the less weight you can take on.

Having smart glasses automagically telling u how much your consuming for each meal & total caloric daily intake via audio or visually within the glasses i think would be powerful! Majority of people have no idea how much they consume and I'm betting they consume anywhere between 2000 to 6000 (heavier people) calories.

I think you are missing the point. I am saying that a smart glass or AI will be unable to tell the amount of calories with any reasonable level of accuracy because fats (oil for example) are the most calorie dense while also the most invisible in pictures. And even a tiny bit of inaccuracy in this measurement is enough to put one from a calorie deficit (losing weight) to calorie surplus (gaining weight).

Guess you are talking about eating at friends for dinner as all chain restaurants offer calorie counts and im betting up to 20% of all restaurants everywhere offer nutritional info too. If less people choose not to opt at the other 80% more will follow. If a family member cooks it you surely can know how it was prepared if you arent preparing.

As for and noted I eat out always a health-ish chains. The AI is getting the info directly from their sites, so taking a pic the AI knowing my location/what restaurant and matching it with pics restaurants post ..it will be seamless vs. having to now pull out my phone and tell GPT or take a pic.