“Now with LLMs, a bunch of the perceived high quality is determined by your immediate. So you have got customers which might be prompting with totally different expertise or totally different stage of expertise. And the result of that immediate could also be perceived as low high quality, however that’s one thing that’s actually laborious to manage.”
Loïc Houssier, VP Product at Superhuman, shared this attitude on a latest podcast.
AI merchandise differ from basic software program in that the expertise is largely decided by the consumer.
Software program has at all times had a studying curve; grasp Photoshop, for instance, and you may apply Bezier curves constantly, similar to another expert consumer.
AI merchandise promoting outcomes, function in another way. The best output isn’t an an identical fastened output achievable by all expert customers.
As an alternative, it’s a collaboration the place skilled prompts can result in a spectrum of legitimate outcomes primarily based on nuanced intent and context.
How can product groups handle this? They’ll rewrite the consumer immediate – many are – to broaden on the consumer intent and steer a primary question right into a extra nuanced & in the end profitable reply.
Even then, anticipating how a consumer may wish to steer the AI is difficult.
One product method I’ve discovered very helpful is a collection of observe up questions. ChatGPT does this very effectively – like on this instance above, asking for refinement on a broad question.
Similar to a colleague asking for readability, the AI seeks steering. Extra than simply asking for better perception, the questions assist me perceive my request higher.
