this post was submitted on 08 Mar 2024
72 points (86.7% liked)
Technology
59148 readers
2721 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
"Prompt engineering" is simply the skill of knowing how to correctly ask for the thing that you want. Given that this is something that is in rare supply even when interacting with other humans, I don't see this going away until we're well past AGI and into ASI.
Human experts often say things like "customers say X, they probably mean they want Y and Z" purely based on their experience of dealing with people in some field for a long time.
That is something that can be learned. Follow-up questions can be asked to clarify (or even doubts - "are you sure you don't mean Y instead?"). Etc. Not that complicated.
(Could be why OpenAI chooses to degrade the experience so much when you disable chat history and training in ChatGPT ๐)
Today's LLMs have other quirks, like adding certain words can help even if they don't change the meaning that much, etc., but that's not some magic either.
Sure - an LLM can help catch some of those situations. But if anything it makes prompt engineering even more important.
Sometimes the customer actually wants X, and a prompt engineer needs to predict this issue and disable the Y/Z behaviour. Prompt engineering is changing, but it's not going away.
But why couldn't an AI do the same?
Why are you assuming it can never get good enough to correctly figure out the intent and find the best possible response it is capable of?
Sure, it's not there today, but this doesn't seem like some insurmountable challenge.