APPLE STUDY FINDS LARGE LANGUAGE MODELS BENEFIT FROM CLASSIC “CHAINING” PRODUCTIVITY TRICK

Apple has published new research suggesting that large language models (LLMs)—the same kind of systems that power AI assistants like ChatGPT, Gemini, and Apple's own in-house models—can see measurable performance improvements when applied with a decades-old productivity method: breaking down big tasks into smaller, structured steps.

- 

The study, conducted by researchers at Apple's Machine Intelligence Research division and released this week on the company’s Al research portal, examined how prompting techniques inspired by human productivity strategies can improve both the accuracy and reliability of modern LLMs.

AN OLD HUMAN TRICK, NEW AI CONTEXT

At the heart of the findings is a technique long familiar to knowledge workers: dividing complex goals into manageable chunks. In Al, this approach translates into “task chaining” or “step decomposition,” where a model is guided through intermediate steps before producing a final answer.

Apple researchers found that LLMs consistently performed better when encouraged to “think in steps” rather than respond in a single, compressed output. “What has been a well-established productivity principle for humans—breaking down large tasks into smaller units—also appears to enhance the reasoning abilities of language models,” the study noted.

TESTING STEP-BY-STEP PROMPTS