AI Luminary Karpathy Predicts 99% Attention Shift to LLMs, Calls for "Research Apps"

Andrej Karpathy, a leading voice in artificial intelligence and founder of Eureka Labs, recently ignited discussion on social media by asserting that the vast majority of attention in the future will be directed by Large Language Models (LLMs) rather than humans. Karpathy, formerly a key figure at OpenAI and Tesla, proposed a radical rethinking of scientific communication, suggesting the need for a new "research app" format specifically designed for AI consumption. This vision challenges the traditional PDF-based publishing model.

"I often rant about how 99% of attention is about to be LLM attention instead of human attention," Karpathy stated in his tweet. This highlights a fundamental shift in how information will be processed and utilized. In AI, "attention" refers to a mechanism within neural networks, particularly Transformer models, that allows them to weigh the importance of different parts of their input data, mimicking human selective focus.

The concept of a "research app" for LLMs moves beyond static documents. Such a format would likely involve highly structured data, executable code, interactive visualizations, and direct API access, enabling AI systems to directly ingest, analyze, and build upon research findings without the need for human interpretation of unstructured text. This contrasts sharply with the current standard of research papers, which are primarily optimized for human readability.

"What does a research paper look like for an LLM instead of a human? It’s definitely not a pdf," Karpathy emphasized, underscoring the limitations of current formats for AI. This paradigm shift could accelerate scientific discovery by allowing AI to synthesize vast amounts of information, identify patterns, and generate new hypotheses at unprecedented speeds.

The development of AI-native research formats presents both opportunities and challenges. While it promises to unlock new frontiers in AI-driven research, it also necessitates new standards for data representation, reproducibility, and the ethical implications of AI-generated insights. Karpathy's call points towards a future where the primary audience for scientific output may increasingly be intelligent machines.