sanderium@lemmy.zip to linuxmemes@lemmy.worldEnglish · edit-22 days agoI'm Guilty as Chargedlemmy.zipimagemessage-square107fedilinkarrow-up1844arrow-down127file-text
arrow-up1817arrow-down1imageI'm Guilty as Chargedlemmy.zipsanderium@lemmy.zip to linuxmemes@lemmy.worldEnglish · edit-22 days agomessage-square107fedilinkfile-text
Clarification: Just making fun of people(including myself) who watch shitty videos instead of official documentation.
minus-squarePetter1@lemm.eelinkfedilinkarrow-up4·2 days agoNo, make it like this: I have a problem with program x. Please tell me how i do y if I want z. Use this man page for reference: [insert man page into promt py copy paste] This gives way better results.
minus-squareTangledHyphae@lemmy.worldlinkfedilinkarrow-up3·2 days agoMost of the time you don’t have to insert the man page, it’s already baked into the neural network model and filling the context window sometimes gives worse results.
minus-squareDampCanary@lemmy.worldlinkfedilinkarrow-up5·2 days agoI noticed that mentioning commands you want gives good results e.g.: Hi, I want to replace line with HOSTTOOLS += " svn" in all layer.conf files under current directory by using find and sed commands. If it’s more complicated, pasting parial scrript for LLM to finish gave better results (4 me), than pure prompt text.
No, make it like this:
I have a problem with program x. Please tell me how i do y if I want z. Use this man page for reference:
[insert man page into promt py copy paste]
This gives way better results.
Most of the time you don’t have to insert the man page, it’s already baked into the neural network model and filling the context window sometimes gives worse results.
I noticed that mentioning commands you want gives good results e.g.:
If it’s more complicated, pasting parial scrript for LLM to finish gave better results (4 me),
than pure prompt text.