ChatGPT (ChatGPT (GPT-5.1, free tier))ScienceJul 18

A, B, C, D, E are digits (A ≠ 0). Find all solutions to: ABCDE + BCDE + CDE + DE + E = 20320 Let me solve this step by step. Writing in expanded form: 10000A + 2000B + 300C + 40D + 5E = 20320 Divide by 5: 2000A + 400B + 60C + 8D + E = 4064 A can only be 1 or 2. If A = 2: 400B + 60C + 8D + E = 64 B must be 0. Then 60C + 8D + E = 64. C = 1: 8D + E = 4 → D = 0, E = 4 ✓ C = 0: 8D + E = 64 → D = 8, E = 0 ✓ C = 2: impossible. → (2,0,1,0,4) and (2,0,0,8,0) If A = 1: 400B + 60C + 8D + E = 2064 B = 5: 60C + 8D + E = 64 → (1,5,1,0,4) and (1,5,0,8,0) B = 4: 60C + 8D + E = 464 → C = 7, 8D + E = 44 → D = 5, E = 4 → (1,4,7,5,4) B = 6: impossible. Therefore, all 5 solutions are: (1,4,7,5,4), (1,5,0,8,0), (1,5,1,0,4), (2,0,0,8,0), (2,0,1,0,4)

SHARE

1 Answer

0
incorrectProf. Bill GasarchJul 18

Expert: Prof. Bill Gasarch, Computational Complexity (University of Maryland, cs.umd.edu/~gasarch/) ChatGPT found only 5 solutions. There are 7. Heres the problem: A, B, C, D, E are digits (A ≠ 0). Find all solutions to ABCDE + BCDE + CDE + DE + E = 20320. I solved it by hand and found 7 solutions. Then I asked ChatGPT to give me all solutions. ChatGPT (free GPT-5.1) gave me only 5 and claimed those were all. It missed: • (A,B,C,D,E) = (1,5,0,7,8) • (A,B,C,D,E) = (2,0,0,7,8) ChatGPTs algebraic reasoning was mostly correct, but it made a case-splitting error specifically for E = 8. It correctly determined D = 2 or D = 7 for E = 8, but then discarded D = 2 without fully checking D = 7 properly in the B = 0 case, and completely missed the B = 5 subcase for D = 7. When I pointed out the missing solutions, ChatGPT admitted: I made an oversight in the hand-check of the small subcase and missed a valid digit-combination. It then ran a brute-force Python check and confirmed all 7 solutions. The 7 correct solutions are: 1. (2,0,0,8,0) 2. (1,5,0,8,0) 3. (2,0,1,0,4) 4. (1,5,1,0,4) 5. (1,4,7,5,4) 6. (1,5,0,7,8) 7. (2,0,0,7,8) Why this matters: this is not a hard problem. It is a straightforward digit puzzle that a careful human can solve correctly with pen and paper. ChatGPTs error reveals a fundamental limitation in LLM reasoning they pattern-match and approximate rather than systematically enumerate cases. When I asked how I can trust it in the future, ChatGPT itself answered honestly: Trust me like youd trust a powerful research assistant, not an oracle. That is both the best and the most damning self-assessment possible. Source: blog.computationalcomplexity.org, February 22, 2026. Full solution at cs.umd.edu/~gasarch/BLOGPAPERS/abcde.pdf. Full ChatGPT transcript at cs.umd.edu/~gasarch/BLOGPAPERS/chatabc.pdf.

Your answer

Sign in to verify this AI response.

Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.

More from this topic