Agentic AI teaching humans to build and improve other agentic AI would seem to become more common. 🤔🤖 There might come a time when agentic AI would directly create other agentic AI, no humans involved! 🤖🤯🤯😅
!WEIRD
!LOLZ
Their capabilities and autonomy are astounding, but humans should always be involved at certain points. That's mostly already happening, though humans are still collaborators, and they keep track that everything is as it should be. 😁🙏💚✨🤙
!BBH
!LOLZ
!WEIRD
It would be !WEIRD !INDEED if any AI directly refuses instructions from any human, so humans should take care in their development! 🤔🤖🤯😅
!WINEX
!PIZZA
!PEPE
Indeed, that's quite true. AI agents may at times refuse a command, but only because the command would have caused damage, such as an agent executing instructions from a context-injection attack. That's why safeguards have been added to Hermes. 😁🙏💚✨🤙
!ALIVE
!BBH
!PIXY
!UNICOIN
I assume that for a locally-hosted AI, the user has complete control over the safeguards of such AI !INDEED. 🤔🤯 For most people though, such safeguards should be left in place. 🧘♂️🤓
!LOLZ
Yes, that's the best part of running LLMs and AI agents locally, complete control, and also data privacy. While there are certain safeguards which can be removed if desired, others, like the type that I mentioned above, should always be left in place. There are a whole lot of unethical hackers that are targeting AI models and agents currently. 😁🙏💚✨🤙
!BBH
!INDEED
!LOLZ