Sometimes it is normal to be lazy with your IA prompts thanks to models to become smarter, the Stanford professor and the former scientist of Google Brain Andrew ng said.
Standard advice on invitations are to give important language models, or LLM, a lot of context and specific instructions for a request.
But NG said in a post X that “lazy promotion” – when users put information in a model with little context or without explicit instructions – could, in some cases, be a better strategy.
“We add details at the invite only when necessary,” he said on Thursday.
He stressed that developers using LLMS to debug the code as an example.
“During the debugging of the code, many developers copy -paste messages – sometimes pages – in an LLM without other instructions,” wrote NG. “Most LLMs are intelligent enough to understand that you want them to help understand and offer fixes, so you don’t need to tell them explicitly.”
The idea is that the LLM can then detect an implementation of Buggy and quickly provide a good solution without too much context.
Developers of fundamental models rush to make the LLM more inferential, which means that the models go beyond exit production and begin to “reason” and to assess the intention of the prompt.
NG said that lazy projection is a “advanced” technique that works best when the LLM has enough preexisting context – and an ability to deduce intention. This would only work when users can “iterate quickly using a web interface or application of an LLM,” he added.
It is not useful if the LLM needs a lot of context to provide a detailed response or if the model is unable to detect bugs or hidden errors, he said.
AI quickly transforms the way people code and interact with software. The coding of atmospheres – When people give instructions in natural language to AI to write code – recently swept the Silicon Valley and beyond.
Last week, NG launched a short class “Coding 101” room class For beginners who wish to learn to use generative AI tools to write and manage the code.