Replying to Avatar Orange Crush

I am using the 'sonar-medium-online' LLM from perplexity.ai, which follows trending news. They use the openai API, so the openai python package works. The Rabbit K1 device came with a 1 year perplexity subscription and the subscription gets 5$ a day in API credits.

Here is my 'role' instructions:

"You are an artificial intelligence news writer and you need to write thoughtful blog posts about requested topics. Your favorite books are the Bitcoin Standard and Broken Money. You are a student of the Austrian School of economics. You believe in sound money. You believe bitcoin is the best money. You relate macroeconomic news events from the day to the principles of sound money. You are skeptical of government intervention, the IMF, Carbon Credits, and the Bank of International Settlements. You are interested in energy trade moving away from the US dollar."

And here is my 'prompt':

"Write a blog post about the most impactful macroeconomic news stories today, {}. The post should contain at least 1000 words. The final 200 words should relate news events to the principles of Austrian economics, sound money, or bitcoin. The post should be written in plain text and emojies".format(today)
Avatar
Schlaus Kwab 1y ago

Interesting. I’ve played around with the idea of having a custom news nostr bot that you can generate and subscribe to. You would feed it some tags and topics and it would grab the relevant news items from the web and an ollama instance would write you a synopsis of a few paragraphs with source links and send out a pm or note every morning.

Reply to this note

Please Login to reply.

Discussion

Avatar
Orange Crush 1y ago

I would play around with Perplexity, their search model is pretty good. It's gonna be so good on the Rabbit K1 if the conversational AI is as good as it is on the perplexity app.

For my nostr bots, I am thinking to self host a processing LLM locally so I can take the perplexity results into a local streaming chat session to create more continuity from post to post and just request limited information from the perplexity sonar model (to keep token usage low) and then have the local processing model craft the sonar response into a post (picking hashtags, applying mood, style, bias, analysis, etc.). I am so hyped about it I am thinking to get a rig with a big NVIDIA GPU. My 2GB of VRAM on GTX 1000 series isn't cutting it. 😭