Announcing our new Paper: The Prompt Report, with Co-authors from OpenAI & Microsoft!
Check it out →We have seen a number of intermediate promting techniques thus far. In this section we will cover advanced methods where an LLMs communicates with other tools to solve complex reasoning tasks. Though this is a relatively new area of research, it has driven significant innovations in prompting techniques. These methods expand the scope of problems that we can solve with prompting. By performing searches via the internet, queries to an external calculator, or sending code to be run elsewhere, the LLM is able to add information that it was not trained on into it's context.
Oftentimes, these methods have been necessitated by shortcomings of LLMs in areas such as mathematical
computations, reasoning, and factuality1. For example, when faced with a question such as What is 19 percent of 5619?
an LLM may fail to provide an accurate answer. To simplify things, the LLM could decide not to provide an answer,
and instead, call upon a calculator tool. A response may look like the following:
What is 19 percent of 5619?
CALCULATOR[(0.19) * (5619)]
In the above example, instead of having to generate an answer out of box, the LLM is able to just provide a format for the answer, while the calculator tool takes care of the computation. This programmatically offloads a task that is difficult for an LLM onto a tool designed for the task. Thus, it is easy to see how such techniques can be useful for the development of GenAI agents.
Though we have provided a simple example, these techniques reach a new level of complexity when API calls, code and reasoning are involed. Some of the methods we will go through are MRKL Systems1, ReAct2 and PAL3 but there are sure to be many more.
Karpas, E., Abend, O., Belinkov, Y., Lenz, B., Lieber, O., Ratner, N., Shoham, Y., Bata, H., Levine, Y., Leyton-Brown, K., Muhlgay, D., Rozen, N., Schwartz, E., Shachaf, G., Shalev-Shwartz, S., Shashua, A., & Tenenholtz, M. (2022). ↩ ↩2
Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., & Cao, Y. (2022). ↩
Gao, L., Madaan, A., Zhou, S., Alon, U., Liu, P., Yang, Y., Callan, J., & Neubig, G. (2022). ↩
Sign up and get the latest AI news, prompts, and tools.
Join 30,000+ readers from companies like OpenAI, Microsoft, Google, Meta and more!