Unstructured Input
Unstructured Input refers to raw, non-formatted data (text, audio, video, code) provided to large-language-models without rigid schema constraints, relying on the model’s inherent comprehension capabilities rather than explicit structural instructions.
Key Developments
- Prompting 2.0 Paradigm: Andrej Karpathy introduced a shift from traditional structured prompting to leveraging LLMs’ ability to comprehend unstructured context directly, achieving advanced results with minimal instruction overhead Karpathy’s Prompting 2.0: Leveraging Unstructured Input for Advanced LLM Comprehension.
- Viral Adoption: The technique gained significant traction via the video “Andrej Karpathy just changed how he prompts claude]
- Leaner Prompts & External Context: Recent shifts by Anthropic and OpenAI emphasize using leaner prompts and external context over complex internal instructions, further reducing the need for rigid schema constraints Anthropic and OpenAI’s New Prompting Rules: Leaner Prompts and External Context.