I provided this input to GPTv:

Here are the results: https://pastebin.com/1eQEUjif

It made many mistakes!

1. OJ cost $6, not $4
2. Tea cost $4, not $10
3. Potatoes were double counted
4. The costs of each person should be divided by the subtotal, not the balance
5. Network error!
6. (?) Ideally it would even point out that the potatoes are probably a side and not a single person's meal, but not a requirement.

I haven't checked the math thoroughly, there might be more. When will I be able to do something like this and have it reliably give the correct answer?

Fine Print
I'll use my own judgment for reliably, probably uploading 3-4 receipts to check.

I should be able to give arbitrary natural language instructions like "The garlic bread was split by the people who got X and Y" or "It's Z's birthday, so split their food evenly among everyone". It should take an image as input - I shouldn't have to type everything out myself.

The tool must be available to me and it should be under $30/month in 2023 dollars, or be more that than but I decide to pay for it anyway (I would not pay more than that for GPTv as it currently exists, FWIW).

• Update 2025-07-05 (PST) (AI summary of creator comment): In response to a user's question about what is needed for resolution, the creator has clarified their process:

  • A candidate tool must be able to solve the specific receipt problem from the market description. The creator has confirmed a tool has passed this test.

  • Final resolution is pending the creator testing the tool's reliability on a few more receipts.