Understanding graph of ideas prompting and a number of other variants…

22 min learn

·

11 hours in the past

Superior prompting methods like chain of thought [8] and tree of thought [9] prompting have drastically improved the flexibility of huge language fashions (LLMs) to resolve advanced, reasoning-based duties. At a excessive stage, forcing the LLM to assemble a step-by-step response to an issue drastically improves its problem-solving capabilities. Nonetheless, all of such methods assume that the reasoning course of ought to observe a linear patterns that progresses from one thought to the subsequent. Notably, the reasoning course of adopted by people tends to be fairly totally different, following a number of totally different chains of thought and even combining insights from totally different ideas to reach at a ultimate answer. Inside this overview, we shall be finding out a number of prompting methods that mannequin the reasoning course of as a graph construction — relatively than a series or tree — that higher captures the assorted forms of non-linear patterns that will happen when reasoning over an issue.

“Human pondering is usually characterised by its skill to make sudden leaps and connections between seemingly unrelated concepts, which may result in novel insights and options. This non-linear, leaping thought course of is a trademark of human creativity, reasoning, and problem-solving skills.” — from [1]

Inside this overview, we are going to discover a number of superior prompting methods for LLMs that can be utilized to resolve tough multi-step reasoning issues. Fortunately, we have now not too long ago overviewed the essential concepts behind prompting, together with:

• Prompting fundamentals (i.e., immediate engineering, context home windows, construction of a immediate, and many others.) [link]
• Superior prompting methods (e.g., chain of thought, self-consistency, and least-to-most prompting) [link]

We’ve lined each sensible and superior prompting methods prior to now. All of those ideas — particularly chain of thought (CoT) prompting [8], self-consistency [10], and tree of thought (ToT) prompting [9] — shall be related for gaining an understanding of this overview. Past these concepts, we have to perceive the transformer structure and the graph…