General idea:
[Analyse Posts]->[Feed through small LLM]->replace original post text with
summarised version(need to have some click through to actually get the
original but default to the summarised version)
In theory a <10b model should be enough for this, will have to test this
out. If I can get away with a smaller model I should be able to rely on
quantised versions running on the actual device ratheere than relying on
cloud hosted models.
Prompt should ideally emphesise retaining company/people names. Shouldn't
have to worry about reading threads for now but that might be the case if
expanded to other paltforms.
SYSTEM PROMPT: You are a summariser model, your task is to summarise posts
from the LinkedIn platform, you will be provided posts in the format
text and should respond in the format summaried
version of original post text Respond only in the format
described, do not provide any additional response or commentary. Ensure
any mentions of company or people names are retained in the summary.