Large Language Models (LLMs) have revolutionized how we interact with technology, but they've always had a bit of an Achilles' heel: multi-step reasoning. Think complex math problems or intricate logical deductions – areas where LLMs often stumble. Why? Traditional methods for feeding LLMs examples (called "in-context learning") rely heavily on text similarity. But, as any math student knows, two problems can look very different on the surface yet share the same underlying logic. This is where GraphIC comes in. This innovative approach ditches text comparisons and instead represents reasoning as graphs, mirroring the way humans think. Imagine mapping out the steps of a math problem, connecting each idea to the next. That's essentially what GraphIC does, creating "thought graphs" for both the problem and potential examples. This clever technique allows GraphIC to find examples that truly align with the problem’s logic, even if they don't share similar wording. Using a probabilistic model inspired by the human brain, GraphIC then selects the best examples to guide the LLM's reasoning process. The results? Across a range of complex tasks – from mathematical reasoning to code generation and logical deduction – GraphIC supercharges LLMs, helping them tackle problems they previously struggled with. GraphIC not only boosts performance, it also offers a peek inside the LLM's "mind," making its reasoning more transparent and interpretable. This is a big step forward, paving the way for more robust and reliable LLMs capable of handling the complexities of human thought.