3 ms·
I’m not a proompter but why can’t LLMs have an initial step to generate an effective prompt from user input, clarifying if intent is not clear enough, and then
by presentation 2y ago
I’m not a proompter but why can’t LLMs have an initial step to generate an effective prompt from user input, clarifying if intent is not clear enough, and then feed that prompt back to the LLM to better fulfill the user’s request? Seems like people have been busy generating training data on what an effective prompt looks like, and kind of silly to require people to learn some verbal gymnastics when the LLM itself is specialized at writing in specific styles.
- nestorD 2y agoIt is already done by some platforms (for example ChatGPT generates a prompt based on your query when generating images). The downside is that you might lose information (the improved prompt might not be an exact semantic match with the original prompt), plus it becomes less and less useful as model's ability to understand commands improve. Basically, if you express yourself clearly, betting on model improvement is enough, there is no need to add a faillible extra-step.
- antonvs 2y agoAnthropic just released a prompt generator this month: https://docs.anthropic.com/en/docs/prompt-generator https://docs.anthropic.com/en/docs/prompt-generator Don’t underestimate how bleeding edge this all is.
- jamesponddotco 2y agoI created most of my prompts[1] using that generator and manual work. I don't think I ever had prompts that worked as well as those. [1]: https://sr.ht/~jamesponddotco/llm-prompts/ https://sr.ht/~jamesponddotco/llm-prompts/
- perelin 2y agoThese guys came up a with a "preprompt" for ChatGPT that does a decent job in refining your initial prompt and asking questions for clarification. Seems they mix some of the best practices like chain of thought, etc. https://github.com/ProfSynapse/Synapse_CoR https://github.com/ProfSynapse/Synapse_CoR
- arthurcolle 2y agoThis is meta prompting