 | Spending some time digging into the system prompts behind agents like v0, Manus, ChatGPT 4o, (...) It's pretty interesting seeing the common threads emerge – how they define the agent's role, structure complex instructions, handle tool use (often very explicitly), encourage step-by-step planning, and bake in safety rules. Seems like a kind of 'convergent evolution' in prompt design for getting these things to actually work reliably. Wrote up a more detailed breakdown with examples from the repo if anyone's interested in this stuff: https://github.com/dontriskit/awesome-ai-system-prompts Might be useful if you're building agents or just curious about the 'ghost in the machine'. Curious what patterns others are finding indispensable? [link] [comments] |
Predictive technology: computational social science
![[link]](https://i.redd.it/36hdamkw77ue1.png){kind=link}