Can AI Boost High School Math Problem Solving? A DeepSeek Case Study

DeepSeek's Reasoning Highlights

When solving the 2024 high‑school entrance exam sequence problem, DeepSeek demonstrated notable logical reasoning abilities, though it sometimes wandered before arriving at the correct answer.

1. Strong logical reasoning from multiple angles

For the first sub‑question, DeepSeek clearly analyzed the problem, tried various combinations, and used elimination and construction methods rather than brute‑force search, illustrating valuable mathematical thinking.

2. Deep exploration and self‑correction

In the second sub‑question, DeepSeek repeatedly approached the correct conclusion but missed a key number, showing a persistent exploratory spirit and occasional "aha moments" that resemble genuine thinking.

3. Offering alternative perspectives

For the third sub‑question on probability, DeepSeek proposed a solution different from standard answers, suggesting that AI can provide novel viewpoints even if not always correct.

Below is a snapshot of DeepSeek's initial restatement of the problem:

For the first sub‑question, DeepSeek quickly gave the answer (1,2),(5,6),(1,6) , which is easy to verify.

Its reasoning process, however, involved many repetitions and obvious errors, yet it eventually arrived at a solution, displaying several "aha moments" captured in the following diagram:

Attempting to delete i=1 and j=3 leaves five numbers; this was identified as incorrect because when m=1 the original sequence has six numbers, and deleting two should leave four.

When m≥3, the original sequence has 4m+2 terms. After deleting positions 2 and 13, the remaining positions split into a front part (1,3,4,…,12,14) and a back part (15,…,4m+2). DeepSeek mistakenly omitted 14 in the front part and included it in the back part, leading to a dead‑end until the response length limit was reached.

<code>Group 1: 1,4,7,10

Group 2: 3,6,9,12

Group 3: 5,8,11,14</code>

Proper Ways to Use DeepSeek for Math Learning

1. Use it to train thinking, not as an answer machine

Instead of feeding a problem and copying the answer, first attempt the problem yourself, then ask DeepSeek for explanations, and actively question its derivations.

2. Let DeepSeek brainstorm multiple solution paths

Ask DeepSeek for at least three different methods, have it critique your approach, and request alternative restatements of the problem to deepen understanding.

Ask for ≥3 distinct solutions and compare efficiency

Ask DeepSeek to criticize your solution and expose potential flaws

Ask DeepSeek to restate the problem from different angles

3. Train yourself to audit DeepSeek's reasoning

Since AI can err, use its explanations as a "error‑checking" exercise: identify mistakes, request a revised reasoning, and compare your own methods with its output.

In summary, DeepSeek’s powerful logical capabilities can greatly enhance mathematical reasoning when used correctly, but it should never replace the critical thinking process that genuine problem solving demands.