𝙾.𝙺. 𝙱𝚞𝚝 𝚆𝚑𝚊𝚝 𝚆𝚊𝚜 𝚈𝚘𝚞𝚛 𝙰𝚗𝚜𝚠𝚎𝚛?
(media.scored.co)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (26)
sorted by:
I find that the best way to use AI is to treat it like a very enthusiastic teenage subordinate. Delegate things to it, but assume it will make mistakes, or take shortcuts to preserve its own ego. Give detailed, precise instructions, and be prepared for it to make incorrect assumptions.Verify everything it does, and know that you need to have final say and culpability in everything it does.
It's fast and very useful, but you need some skill at wrangling it.
And, like you said, it's far more powerful as a partner to bounce ideas back and forth with while you do the actual work, than it is when you just tell it to do something.
And yes, AI being implemented into workflows is basically inevitable. Doing it correctly will be the trick.
Also like teenagers, they have no persistent memory of things you taught them! You have to repeat your specifications and rules everytime it cleared its cache, you'd thought no imaginary sources should have been default but Noooo you have to repeatedly tell it not to lie
Exactly. Enthusiastic, but in need of constant direction and guidance.
A big advantage over actual teenagers is they won't get pissy if you tell them to start over and do it all from the top. Which is an important thing to do. Trying again fresh, or pitching the question at a different LLM is a good sanity check when you can't personally verify something that sounds plausible. If it's a hallucination, it will be different each time.