@mattpocockuk I have found recently that if you create a “system” where the LLM “knows about you”, then step 1 almost disappears (CRM, WhatsApp, email, project management, etc.). One shooting everything. Although it could also be Opus 4.7 too good 🤔
@IanTimotheos@GeoffreyHuntley@IanTimotheos I find it better if you say “To make sure we are on the same page, explain to me [what I want][what we will do][why it is relevant]etc.
Forcing it to explain unlocks some big brain analysis! 🥷
@ValerioCapraro The “LLMs don’t really understand” sounds a bit harsh! Models were trained to do this, especially GPT. The question is what does all of this pre-training mean for users 🤔
Papa @badlogicgames is not convinced, but I find experiments with message passing with pi highly insightful. This demos how one pi can send a message to another session and back:
@nasqret Yes, you are right. While the skill does not instruct it, the inference provider decides to search the web to find any public/open data (so captures arxiv papers and similar)
@nasqret I dug into the /octo:research skill to understand how it works, and from what I can see it doesn't query scientific databases or the web, it just sends text prompts to LLM CLIs and collects their responses.
Am I missing something in the implementation? 🤔
@heeney_luke Agent-Based Modelling and Multi-agent systems have been a proven methodology for decades in economics. The real challenge (if used for policy) is validation at the micro (human behaviour🧍) and macro (society 🏙️)
Exciting times 👏