XPost: sci.logic, sci.math   
   From: janburse@fastmail.fm   
      
   Hi,   
      
   With RAG it could react on news. RAG (Retrieval-   
   Augmented Generation) and other techniques can   
   make a local AI evolve dynamically — both in knowledge   
      
   and personality. Lets say I put such a AI Persona in   
   a News Group loop. I could feedback the posts there,   
   to make it also evolve. One can collect community   
      
   reactions (upvotes, replies, sentiment). Or apply automatic   
   evaluation — e.g., sentiment analysis or engagement   
   score. This becomes your fitness function.   
      
   You can even run multiple personas in the same loop   
   Different viewpoints (e.g., “optimist”, “skeptic”,   
   “data scientist”) They interact and debate posts.   
      
   This is ideal for dead open source projects, such as   
   SWI-Prolog, to pretend user participation, very important   
   in our new marketing world that is totally   
      
   engagement focus. The quality(*) of content doesn't count.   
      
   LoL   
      
   Bye   
      
   (*) Same for research paper mills. BTW, ChatGPT suggests   
   me the following tech stack:   
      
   LLM: Ollama (LLaMA 3, Mistral, Gemma, etc.)   
   LangChain / LlamaIndex: context + memory   
   ChromaDB / SQLite: memory store   
   Feedparser / API client: ingest news or group posts   
   Transformers / VADER: sentiment scoring   
   Simple scheduler or cron job: to run the loop daily   
      
   > Hi,   
   >   
   > There are different ways to make an AI Persona.   
   > For example Twitter aka X, has a grok account.   
   > But grok is not trimmed towards a particular   
   >   
   > profile, maybe an alter ego of Elon Musk was   
   > speculated. What could be interesting would be   
   > to go towards character.ai :   
   >   
   > "Character.ai was established in November 2021.   
   > The company's co-founders, Noam Shazeer and   
   > Daniel de Freitas, were both engineers from   
   > Google. They both worked on AI-related projects:   
   > Shazeer was a lead author on a paper that   
   > Business Insider reported in April 2023 "has   
   > been widely cited as key to today's chatbots",   
   > and Freitas was the lead designer of an   
   > experimental AI at Google initially called Meena,   
   > which later became known as LaMDA.   
   > https://en.wikipedia.org/wiki/Character.ai   
   >   
   > Maybe this can be even done with Local AI. By for   
   > example populating a context. Or creating a kind   
   > of LangChain memory. ChatGPT calls it persistent context.   
   >   
   > So yes you could "spam" an AI Archimedes Plutonium   
   > probably from your desktop with an AI Laptop, maybe   
   > already in 2025, or later in 2026.   
   >   
   > bye   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)