<ul><li>Anthropic has developed a technique called circuit tracing to uncover the inner workings of its large language model (LLM) called Claude 3.5 Haiku.</li><li>Circuit tracing is a new technique that allows researchers to track how an AI model processes information step by step, revealing unconventional methods used by LLMs.</li><li>Claude demonstrated the ability to apply abstract concepts across different languages and use unconventional methods to perform mathematical calculations.</li><li>The research findings show that LLMs have more foresight and do not solely rely on predicting words in sequence.</li></ul>

We are finally beginning to understand how LLMs work: No, they don't simply predict word after word

Discover more