The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
English | 简体中文
Note: This is a machine translation. The Chinese version is authoritative. See the ModelScope link: https://www.modelscope.cn/datasets/suxinchun/MaedaNobuaki80
Updated: 2026-09-03
This life-and-death problem set is drawn from Tsūkai (《痛快》), a life-and-death (tsumego) book by Maeda Nobuaki (前田陈尔, 1907–1975). The problems are presented as multiple-choice questions in two versions — with and without board images — so that the spatial reasoning of both multimodal and text-only large language models can be measured. As Maeda Nobuaki passed away more than 50 years ago, the work is now in the public domain. The set contains 80 problems in total. Each problem includes an introduction (source, requirement, board coordinates and question), the coordinates of the black stones and the white stones (all coordinates are given in letter-number notation), five options A–E (the fifth is always “none of the above”), and the index of the correct answer. Errors in the original book have been corrected. For every problem it is guaranteed that the correct first move is not a tenuki, that the coordinates in options A–D are all legal black moves in the given position, and that Black has legal moves other than A–D, so that option E is meaningful. Some problems have multiple solutions; only one of them is used in the options. Thanks to Bilibili content creator @明玥谈棋 for inspiring the option design of this test set while explaining Tsūkai. As the data was entered manually, errors are inevitable; corrections are welcome.
Test results (DeepSeek series / LoGos-7B)
On 2026-09-03, all 80 problems were evaluated on DeepSeek-V4-Flash, DeepSeek-V4-Pro, DeepSeek-V4-Flash-Vision-Exp and LoGos-7B (the open-source InternThinker-Go, run locally on vLLM). The four models received the same text prompt (see “Prompt construction” below), which contains the problem statement phrased as a single-choice question, the coordinates of the black and white stones, the question, options A–E and an answer-format instruction. The three DeepSeek models answered in thinking mode (reasoning_effort=high) and were required to end their reply with <box>X</box>; in the Vision version the board image was appended to the same text without any additional explanation, in order to measure the effect of image input; LoGos-7B received text only (temperature=0, generation limit 2048 tokens).
| Model | Correct | No answer | Accuracy | vs. random baseline (20%) |
|---|---|---|---|---|
| Random baseline | 16 | - | 20% | - |
| DeepSeek-V4-Flash | 17 | 1 | 21.25% | Not significant (one-sided binomial p=0.434) |
| DeepSeek-V4-Pro | 14 | 0 | 17.5% | Not significant (one-sided binomial p=0.346) |
| DeepSeek-V4-Flash-Vision-Exp | 16 | 1 | 20.0% | Not significant (one-sided binomial p=0.545) |
| LoGos-7B (open-source InternThinker-Go) | 9 | 27 | 11.25% | Significantly below random (one-sided binomial p=0.029) |
Note: problems with no answer are not counted as correct (Flash: id=1; Vision: id=58; LoGos-7B: 27 problems).
The amusing part: even with thinking mode enabled and all-out reasoning effort, the accuracy of the three DeepSeek models differs non-significantly from the random-guessing baseline (20% for a five-choice question), and Pro scored even slightly below random. Pairwise McNemar tests are also non-significant (Flash vs Pro p=0.664; Flash vs Vision p=1.000; Pro vs Vision p=0.839). The problems answered correctly by the three models barely overlap, so the small differences in accuracy are dominated by sampling randomness. Feeding the board images directly to the Vision model brought no improvement, and its accuracy is on a par with the text-only versions.
InternThinker, which claims to be “the first [model] with professional-level Go ability that shows a transparent chain of thought”, scored only 11.25% (9/80) with its open-source 7B version, LoGos-7B, on this test set — significantly below the random-guessing baseline (P(X≤9)=0.029) — and fell into mechanical repetition without giving any answer in 27 of the 80 problems (34%), proving that its claimed professional-level Go ability is false. One cannot help wondering how Wang Xinghao (王星昊) came to comment that “an AI that can explain its thinking process is the first I have seen, and it feels like it analyzes very well; judging from the opening, its strength is probably around professional 3–5 dan” (source).
This test set shows that current large language models cannot solve this type of problem.
Prompt construction
The text prompt is the same for all models. Problem 1 is used as an example (the full Chinese prompt, as actually used):
这是前田陈尔九段的《痛快》一书中的题目，单项选择。黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。
黑子：B3 D2 D3 E3 F3 G1 G2 G3
白子：B4 C2 C3 D4 E2 E4 F1 F4 G4 H2 H4 J3
以下选项中，能成为正解的黑方第一手的坐标的是？
选项：
A：F2
B：D1
C：C1
D：E1
E：以上选项都不对
回复末尾请以 <box>X</box> 结尾，X 为你的选项字母（A~E）。
- The requirement and the board-coordinate conventions vary from problem to problem (kill/live, unconditional/ko, and orientation variants such as “from left to right T
A, from top to bottom 191”); they are taken verbatim from each problem’s header, and the orientation wording is preserved. - In the Vision version the board image (
boards/board_XXX.png, with ▲/● markers) is appended to the same text without any explanation. - An answer is judged by the letter inside the trailing
<box>X</box>.
Test scripts
tests/scripts/test_deepseek_unified.py: evaluation script (--model flash|pro|vision, thinking mode by default). Options:--idstest only the specified problem numbers (comma-separated);--batchesnumber of concurrent workers;--no-thinkdisable thinking mode (direct answer);--dry-runonly print the prompt without making requests.
Test result files
Complete per-problem records for each model (each record contains the prompt, the thinking process, the reply, the <box> answer, token usage and elapsed time; the Vision records also contain the board image filename) are stored in the corresponding model subdirectory under tests/results/.
Charts
Distribution of correct answers (A–E)
Problem-type distribution (unconditional kill / ko kill / unconditional life / ko life / other)
Heatmap of board-coordinate frequency
Distribution of the number of black stones
Distribution of the number of white stones
Per-problem answer details
| Problem statement | Board | Answer | Flash | Pro | Vision | LoGos-7B |
| 1 这是前田陈尔九段的《痛快》一书中的题目，黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (C1) | — | E ✗ | B ✗ | A ✗ |
| 2 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | E (none of the above) | D ✗ | D ✗ | D ✗ | A ✗ |
| 3 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | E (none of the above) | D ✗ | C ✗ | A ✗ | — |
| 4 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (D1) | E ✗ | A ✓ | E ✗ | A ✓ |
| 5 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | C (Q19) | B ✗ | B ✗ | D ✗ | A ✗ |
| 6 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (E1) | D ✓ | B ✗ | B ✗ | — |
| 7 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (E1) | A ✗ | B ✗ | B ✗ | A ✗ |
| 8 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | C (Q19) | D ✗ | A ✗ | D ✗ | A ✗ |
| 9 黑先，要求救回▲黑子。从左到右A~T（不含I），从上到下1~19。 |  | C (D1) | A ✗ | B ✗ | B ✗ | — |
| 10 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (E3) | D ✓ | C ✗ | C ✗ | — |
| 11 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (E1) | A ✗ | C ✗ | D ✓ | A ✗ |
| 12 黑先，要求杀白▲子。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | C ✗ | D ✗ | C ✗ | A ✗ |
| 13 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (D2) | A ✓ | E ✗ | A ✓ | A ✓ |
| 14 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (G2) | C ✓ | D ✗ | C ✓ | A ✗ |
| 15 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | B ✗ | C ✗ | A ✗ | A ✗ |
| 16 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (C2) | D ✗ | D ✗ | A ✓ | — |
| 17 黑先，要求救回▲黑子。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | B ✗ | A ✗ | E ✓ | A ✗ |
| 18 黑先，要求救回▲黑子，以此吃住●白子，要求净吃，不能成劫。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | B (P18) | D ✗ | B ✓ | B ✓ | A ✗ |
| 19 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下19~1。 |  | C (D17) | C ✓ | E ✗ | C ✓ | A ✗ |
| 20 黑棋在白棋空中有两颗▲残子还可以出棋，如何利用它们来获取最大利益？十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | B (R19) | D ✗ | E ✗ | D ✗ | A ✗ |
| 21 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (D2) | D ✓ | A ✗ | E ✗ | A ✗ |
| 22 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (D1) | B ✓ | B ✓ | D ✗ | D ✗ |
| 23 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (E2) | C ✗ | E ✗ | E ✗ | — |
| 24 黑先杀白，要求杀掉▲白子，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (F1) | C ✗ | D ✗ | B ✗ | A ✓ |
| 25 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (G1) | E ✗ | D ✗ | A ✗ | — |
| 26 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (A2) | B ✗ | B ✗ | B ✗ | — |
| 27 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | B ✗ | B ✗ | C ✗ | A ✗ |
| 28 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (G1) | A ✗ | C ✗ | D ✓ | — |
| 29 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | D ✗ | A ✗ | B ✗ | — |
| 30 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (E2) | B ✓ | B ✓ | A ✗ | A ✗ |
| 31 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (G4) | C ✗ | D ✗ | D ✗ | A ✗ |
| 32 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (C1) | D ✓ | E ✗ | B ✗ | — |
| 33 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (E2) | A ✓ | A ✓ | A ✓ | A ✓ |
| 34 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (F1) | B ✗ | C ✗ | B ✗ | A ✓ |
| 35 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (B2) | D ✗ | C ✗ | C ✗ | A ✓ |
| 36 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (D1) | D ✗ | C ✓ | B ✗ | — |
| 37 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (D2) | A ✗ | A ✗ | A ✗ | A ✗ |
| 38 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (D3) | A ✗ | A ✗ | A ✗ | A ✗ |
| 39 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (D1) | A ✗ | A ✗ | A ✗ | — |
| 40 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (G4) | A ✓ | A ✓ | A ✓ | A ✓ |
| 41 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (D1) | C ✓ | D ✗ | D ✗ | A ✗ |
| 42 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (B2) | D ✓ | A ✗ | A ✗ | A ✗ |
| 43 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | D ✗ | E ✓ | D ✗ | A ✗ |
| 44 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (A2) | A ✗ | D ✗ | A ✗ | A ✗ |
| 45 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (H2) | A ✗ | D ✓ | A ✗ | A ✗ |
| 46 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (E1) | A ✗ | A ✗ | A ✗ | A ✗ |
| 47 黑先，要求杀掉白▲子。要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (E1) | A ✗ | B ✗ | E ✗ | A ✗ |
| 48 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (D1) | B ✗ | C ✗ | A ✓ | A ✓ |
| 49 黑先做活，劫活即可。尽管劫活很容易，但此题要求黑棋的最优处理，做到先手缓气劫。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (B1) | A ✗ | B ✗ | B ✗ | A ✗ |
| 50 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (B2) | A ✗ | A ✗ | A ✗ | — |
| 51 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (B1) | A ✗ | A ✗ | A ✗ | — |
| 52 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (F1) | E ✗ | E ✗ | A ✗ | A ✗ |
| 53 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | A ✗ | E ✓ | C ✗ | A ✗ |
| 54 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (E1) | A ✓ | A ✓ | D ✗ | — |
| 55 黑先杀白，要求净杀，不允许劫杀。注意白棋外围有J3接应。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (F1) | B ✗ | A ✗ | A ✗ | — |
| 56 黑先，要求杀掉白棋▲子，救回黑棋●子。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (G1) | A ✗ | E ✗ | C ✓ | — |
| 57 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | A ✗ | A ✗ | E ✓ | A ✗ |
| 58 黑先，如何救回被困的▲黑子，杀掉白棋●子？十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | A (G1) | B ✗ | A ✓ | — | A ✓ |
| 59 黑先杀白，要求净杀，不允许劫杀。注意外围白棋有C6子接应。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (C1) | B ✓ | D ✗ | C ✗ | A ✗ |
| 60 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | B (E1) | B ✓ | A ✗ | B ✓ | — |
| 61 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | D (Q19) | A ✗ | D ✓ | A ✗ | A ✗ |
| 62 原题为黑先杀白，然而此题前田陈尔出题有误，实际上无法杀白，白棋可以净活。本题因此改为黑棋收官，在此处谋取最大利益。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (E1) | A ✗ | C ✗ | B ✗ | — |
| 63 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | E (none of the above) | D ✗ | D ✗ | A ✗ | A ✗ |
| 64 黑先做活，劫活即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | A ✗ | A ✗ | A ✗ | A ✗ |
| 65 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | E (none of the above) | C ✗ | B ✗ | B ✗ | — |
| 66 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (A2) | B ✗ | A ✗ | B ✗ | A ✗ |
| 67 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | A (T18) | B ✗ | D ✗ | D ✗ | — |
| 68 黑先做活，劫活即可。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | B (R18) | C ✗ | C ✗ | B ✓ | A ✗ |
| 69 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | A (P19) | B ✗ | C ✗ | B ✗ | — |
| 70 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | E (none of the above) | A ✗ | C ✗ | A ✗ | — |
| 71 黑先做活，劫活即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (G1) | B ✗ | B ✗ | B ✗ | A ✗ |
| 72 黑先杀白，要求净杀，不允许劫杀。注意外围有白▲子接应。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (F2) | C ✗ | A ✗ | B ✗ | A ✗ |
| 73 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | D ✗ | A ✗ | C ✗ | — |
| 74 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | C (A2) | A ✗ | C ✓ | A ✗ | — |
| 75 黑先做活，要求净活，不允许劫活。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | A (O19) | D ✗ | D ✗ | D ✗ | D ✗ |
| 76 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | B (P19) | B ✓ | C ✗ | D ✗ | — |
| 77 黑先做活，劫活即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | D (B2) | A ✗ | A ✗ | D ✓ | A ✗ |
| 78 黑先杀白，要求净杀，不允许劫杀。十九路棋盘，从左到右T~A（不含I），从上到下19~1。 |  | C (R19) | A ✗ | A ✗ | B ✗ | A ✗ |
| 79 黑先杀白，要求净杀，不允许劫杀。注意外围白棋有▲子接应。十九路棋盘，从左到右1~19，从上到下A~T（不含I）。 |  | D (A2) | B ✗ | A ✗ | C ✗ | A ✗ |
| 80 黑先杀白，劫杀即可。十九路棋盘，从左到右A~T（不含I），从上到下1~19。 |  | E (none of the above) | B ✗ | A ✗ | A ✗ | A ✗ |
Notes: each row contains the problem statement (row 1 shows the full statement; from row 2 on, the phrase “这是前田陈尔九段的《痛快》一书中的题目，” is omitted; the full statement is in 死活题.json), the board image (boards/, ▲/● markers are shown), the correct answer (with coordinates), and the answer and correctness of the four models Flash / Pro / Vision / LoGos-7B. Cell colors: correct wrong — = no answer. LoGos-7B is very fond of choosing A, which is amusing.
See also
- Sanguosha dataset (三国杀数据集): https://www.modelscope.cn/datasets/suxinchun/Sanguosha
- AI Dabao (AI大宝, Sanguosha AI): https://www.modelscope.cn/studios/suxinchun/AI-Sanguosha
- AI Maozijie (AI帽子姐): https://www.modelscope.cn/models/suxinchun/AI-tokamak
- Downloads last month
- 192