Looks like the hardware needed to run state-of-the-art open source LLM will be ~100kg... haha
It's not feasible today. Someday!
8 karma · joined August 9, 2022
Tool calling is very helpfull when humans can manually define those functions. However, it's also very limited because humans have to define those functions manually unless the function is very simple without any external resources such as accessing an external API with API keys, etc.
Someday, we'll have AI agents that work like our secretaries or employees. But just like a CEO doesn't have to know how to use all the apps that his employees use, some apps would be created only for AI agents. In such a scenario, the interface of these apps would look very different from HCI. For example, they most likely wouldn't use vision or sound as a part of the main UI, unlike HCI.
Interestingly, most apps can be re-rewritten as a Decision Tree. If you push this button you'll be navigated to this page, where a different list of actions is available, etc. I think an AI-Computer Interface if it is ever invented, might look like a text-based Decision Tree without vision or sound unless they are absolutely necessary.