Personal Science Week - 250508
Better food tracking with the new "Everything App"
Back in PSWeek230608 I shared my experience with food tracking apps, where I recommended a specific app (Cronometer) along with the advice to just track rigorously for a few weeks and then revert to a more informal note-taking method.
As with most things, LLM chatbots like ChatGPT have become so good that the old advice no longer applies. This week I’ll describe my new way to track diet.
Better Food Tracking
Everyone, personal scientist or not, should track their food consumption at some point. You don’t have to be rigorous. Often the simple act of noting what you ate, and the awareness that results, is enough to uncover useful improvements.
Until recently, my method of food tracking was to set aside a few days or a week with a dedicated app (I prefer Cronometer) and a food scale. Log everything meticulously long enough to get a sense of your typical eating habits. Then stop. Analyze the results, either with the app’s built-in tools, or exported to Excel or your own favorite analysis system.
Not anymore. Recently I simply set up a conversation within one of my LLM chatbots (usually ChatGPT or Claude) and chat away. The voice input is so accurate, and the results so easy to adjust and analyze, that I simply do it on the fly.
Say things like: “It’s 10am and I’m having another latte”, “It’s 3pm and I’m eating a banana”, “One chicken thigh with pesto”.
Then any time during the day I can ask it to generate a plot showing my progress:
That’s it! I no longer subscribe to a dedicated food tracking app. My LLM has become my “everything app”.
So once you’ve settled on an LLM, is there any point to using a dedicated food tracking app? As with everything else of importance, I ran a “Deep Research” investigation to try to answer that question, this one with Gemini looking at over 100 sources in order to compile a 30-page report Specialized Nutrition Applications vs. Conversational AI: An Analysis of Capabilities, Accuracy, and Future Trajectories in Digital Food Tracking. Tldr; specialized food-tracking apps have built-in features that can make rigorous logging more convenient, like one-click adding your favorite foods, and many of them are adding their own AI features, but otherwise most people are better off using an LLM.
Readings in Food Tracking
No surprise to the skeptical personal scientist, but no matter what app you use, it’s important to remember that the nutrition information might not be all that accurate. A New York Times short documentary and editorial highlights the problem: Reporter Casey Neistat took ordinary foods and tested them with a professional calorimeter (the gold standard for measuring caloric content) and discovered significant discrepancies.
In his sample of everyday foods — a Subway sandwich, convenience store muffin, Chipotle burrito, and vegan deli sandwich — the differences between labeled and actual calories added up to more than 500 calories in a single day. That's equivalent to an entire Big Mac or a couple of Snickers bars that don't appear on any nutrition label.
If you really get into food tracking, here are some additional sources to consider:
Nutrition Pro: If you want to create your own FDA-approved nutrition labels, this service costs about $50/month (https://nutritionistpro.com/nutrition-food-labeling-pricing/)
TellSpec: For the dedicated self-tracker, this portable Mass Spectrometer allows direct measurement of food components (https://tellspec.com/order/)
Sift App: Provides detailed breakdowns of ingredients in packaged products. The first 5 scans are free, followed by a subscription of approximately $3/month (https://www.siftfoodlabels.com/
We’ve discussed other related topics, including an app to evaluate how processed your food is, an energy balance calculator, and more in our archives.
As always, the personal science way is to be open-minded but skeptical, and remember to keep a broad margin of error in your calculations. Might these discrepancies help explain why some people find calorie counting less effective than expected?
Weekly Personal Science Readings
Long-time Longevity researcher Matt Kaeberlein says “Bryan Johnson is not credible”
No one should base their health decisions on his claims. His recent statements about rapamycin appear more like a publicity stunt to shift attention away from negative press than anything grounded in data or science.
If you’re a personal scientist looking for your way to make your mark on the world, Convergent Research’s “Fundamental Development Gap Map 1.0” is a list of science problems that need solving. Like “Limited Microbial Hosts/Chassis Organisms”: how come everybody uses the same ole E. Coli as the model organism? Wouldn’t we learn more if we could dedicate similar in-depth modeling to some other microbes?
Wm Briggs on why p-values are a bad way to do science: If your knowledge of statistics comes from a long-ago class in college or from what you’ve read and learned over the years, you may vaguely think of of p values as a quick-and-dirty measure of statistical significance. But if you’re serious about truth, relying on p<0.05 (the standard rule of thumb) is a bad idea.
Two recent tips from one of our favorite Substacks: Recomendo:
The small, lightest CO2 monitor is AirSpot ($144), which is smaller than the size of a thumb drive, and will give you instant CO2 levels. I carry it in my pocket when travelling; if the levels get high, I can choose to mask, or exit if possible. (The highest level I’ve seen so far is in the waiting room at the DMV.) — KK
and
This article lists 20 verification tools for combating misinformation. Included are fact-checking sites like Snopes and Google’s Fact Check Explorer, as well as reverse image search engines, identity verification sites, and AI detection tools.
About Personal Science
Personal science is about bringing the rigor and methods of scientific inquiry to our everyday lives. Unlike professional science, which is driven by academic interests or corporate profit, personal science is motivated by individual curiosity and the desire to optimize our own health, productivity, and wellbeing.
If you have other topics you’d like to discuss, please let us know.
Even if food was labelled with complete accuracy, not everyone absorbs nutrients equally efficiently at all times, and how food is prepared can make a big difference, too!
I see labels less as data, and more as an educational tool...
I tried using chatGPT to design a meal plan for grams of carbs, protein, and fat, but when I checked the math it didn’t add up. Might want to check your diet for a day with Cronometer and see if the LLM can do math.