Content pfp
Content
@
https://warpcast.com/~/channel/aichannel
0 reply
0 recast
0 reaction

Dan Romero pfp
Dan Romero
@dwr.eth
what’s the equivalent of “[query] + site:reddit.com” for LLM prompts?
6 replies
2 recasts
24 reactions

Gabriel Ayuso pfp
Gabriel Ayuso
@gabrielayuso.eth
There are many hacks that can lead the model towards a weight space that has better answers depending on your needs. Role prompting is a common one. "Act as a principal security engineer". Asking for points of view from specific individuals or publications (hallucinations tend to happen more depend on source). Other reasoning related like breaking down by step, or "take a breath" and other such hacks. It's been almost a year since I was deep in this work though so things might've changed.
2 replies
0 recast
7 reactions

Royal pfp
Royal
@royalaid.eth
Let's think step by step has basically evolved into reasoning models as far as I understand it but pushing the model to a better latent space is still an open question AFAIK, it's mostly lore and tricks currently.
1 reply
0 recast
1 reaction

Gabriel Ayuso pfp
Gabriel Ayuso
@gabrielayuso.eth
Yeah reasoning models are basically chains of thought supercharged with loops and other tricks afaik.
0 reply
0 recast
0 reaction