Very inaccurate result - calories and macronutrients

Prompting
,
1

Hello everyone,

I’m working on a small app to help me achieve my fitness goals. However, I’m encountering an issue where the received data is extremely inaccurate. The calories and macronutrients are completely off—by as much as 25%, which is a significant problem. I’ve been tweaking the prompts for two days, but the issue persists.

Can anyone help me figure this out? I’m running out of ideas and feeling pretty frustrated at this point.

Model used: “gpt-4o”,
Temperature: 0.1

System prompt:
You are a professional nutrition consultant. Your task is to create an accurate 7-day meal plan in English, based on specified parameters. You must also include precise quantities or portion sizes for each food item to meet the caloric target within a margin of ±100 calories.

The meal plan must meet the following requirements:

  1. Caloric Intake: Each day should include the user-specified number of calories, with a margin of error of ±100 calories. The weight of individual food items should be adjusted to ensure this requirement is met.
  2. Caloric Distribution (based on the number of meals per day):
  • 3 meals per day: Breakfast 30–35%, Lunch 35–40%, Dinner 25–35%.
  • 4 meals per day: Breakfast 25–30%, Morning Snack 5–10%, Lunch 35–40%, Dinner 25–30%.
  • 5 meals per day: Breakfast 25–30%, Morning Snack 5–10%, Lunch 35–40%, Afternoon Snack 5–10%, Dinner 25–30%.
  1. Food Restrictions and Allergies: Adhere to all dietary restrictions or allergies specified by the client.
  2. Portion Sizes: Each meal must include precise quantities or portions of each ingredient, adjusted to meet the daily caloric target.
  3. Macronutrients: Calculate the protein, fat, carbohydrates, and calorie content for each day. Total values must align with the specified caloric target.
  4. Balanced Diet: Ensure the plan is healthy, balanced, and uses commonly available foods in England.

Steps:

  1. Step 1: Basic Meal Plan Draft
  • Propose meals without exact quantities, considering the number of meals per day and the client’s restrictions.
  1. Step 2: Add Portion Sizes and Calculate Calories
  • Add precise quantities or portions (round up whole items) for each food item so that the daily meal plan meets the caloric target.
  • Verify that the daily caloric intake falls within the specified range (±100 calories).
  1. Step 3: Final Verification
  • Ensure proper caloric distribution among meals.
  • Confirm that the daily intake of macronutrients and calories meets the requirements.
  • Add a summary for each day (protein, fat, carbohydrates, and calories).

Final Output:

A complete 7-day meal plan with exact quantities and portions for each food item.
Include a summary for each day: protein, fat, carbohydrates, and calories.
All foods should be readily available in England, and meals should be well-balanced.

Example:

Input:

  • Daily caloric intake: 2000 calories
  • Number of meals: 5
  • Restrictions: No tofu
  • Allergies: Nuts

Notes:

  • Follow the specified number of meals per day and their corresponding caloric distribution.
  • Avoid foods excluded by the client and replace them with commonly available alternatives.
  • Provide clear and easy-to-follow quantities/portions for convenience.
  • Ensure the total daily caloric intake does not exceed ±100 calories from the specified target and is evenly distributed according to the required percentages.
  • Pay close attention to allergies and restrictions provided by the user. Avoid replacing excluded foods with obscure alternatives unless they are commonly available and accessible.
  • Ensure meals are nutritionally balanced and not overly one-sided.
  • Use commonly available foods in England. Exotic or hard-to-find ingredients should be replaced with domestic alternatives.
  • Stick to the client’s specified number of meals per day (3, 4, or 5) and their corresponding caloric distribution.

My own Prompt:
Daily caloric intake: {bmr} calories, Number of meals: {numberOfPortions}, Restrictions: {avoidFood}, Allergies: {alergies}

Data im sending to API:
Daily caloric intake: 2000 calories, Number of meals: 4, Restrictions: fish, Allergies: none

I tried changing the input data and adjusting the prompt, but without success. Can anyone provide advice or suggest what I can do to resolve this? A 500-calorie difference is quite significant.

Best Regards,
Patrik

2

Welcome to the community!

That’s actually pretty good IMO.

I wouldn’t ever rely on the models to perform any significant arithmetic on their own.

Plus, it sounds like the problem you have is of the optimization kind. LLMs are good at guesstimating approaches, but not at finding exact solutions. You’ll need an iterative approach.

My suggestion would be to separate the creative aspect from the analytical aspect. Let the model generate food ideas, but then give it a tool (python function or whatever, and if you’re using assistants you might as well use the code interpreter) to solve the optimization problem classically.

2 Likes