Zero-shot prompting
In some instances, examples aren’t needed or possible. Asking a mannequin to, say, translate a paragraph into French or Russian isn’t a scenario the place examples make sense. If the mannequin is as much as the duty, the assorted phrases from the opposite language ought to be a part of the coaching set. In these instances, the person can say little as a result of the arduous work has been executed when the mannequin was constructed.
Q&A format
As a substitute of treating the LLM as a minion, the Q&A strategy treats it as a sage with a deep means to reply questions. A collection of brief questions adopted by direct solutions can typically triangulate on the reply, delivering data to the person. This solely works, although, when the LLM is a software for locating data and, maybe, educating. It doesn’t work as nicely when the AI is supposed to generate a block of writing. Some folks really feel that an interview or interrogation constructed from a collection of brief, direct questions reduces hallucinations, however that hasn’t been my expertise. The LLMs are engineered to please us they usually hallucinate after they’re greedy for a solution. Extra questions means extra solutions and extra alternatives to hallucinate.
Socratic prompting
LLMs are famously glib. Some customers don’t need the glib reply, so that they ask the LLM to reply not with a solution, however with a collection of questions concerning the person’s activity or purpose as a substitute. In different phrases, the mannequin is requested to interrogate the human as a check. This strategy is especially standard for fashions which can be deployed in colleges, however it’s useful in any context the place the person desires to achieve understanding or power the mannequin to assume extra deeply about an issue.
