AI/LLM Task Capabilities Cheat Sheet

Large Language Models (LLMs) represent a transformative shift in artificial intelligence, functioning as general-purpose reasoning engines capable of performing hundreds of diverse tasks through natural language interaction. Unlike traditional AI systems narrowly trained for single objectives, modern LLMs exhibit emergent abilities across text generation, multimodal understanding, code synthesis, and complex reasoning — including reasoning models (o1, DeepSeek-R1, Gemini) that scale test-time compute to achieve expert-level performance on scientific, mathematical, and coding benchmarks. Understanding these capabilities matters because choosing the right task formulation directly determines success — the same model can excel or fail based entirely on how you frame the problem, structure the prompt, and select the appropriate inference pattern. A critical insight: LLMs don't execute tasks deterministically like classical programs; they generate probabilistic responses shaped by training data, prompting techniques, and contextual grounding, making reproducibility and factual accuracy ongoing challenges that require deliberate mitigation strategies.