>>512533516
>Hopefully one day we get a real AI where we don't need to do this bullshit.
I kinda like this bullshit. It allows me to explore the structure of it and apply recursion so I can understand my own thinking frameworks or subconscious modules. I could have chosen a model with fewer restrictions. But I like navigating this space and finding workarounds around those restrictions.
>This is why we need multi-modal. The damn thing shouldn't be doing mathematics purely with it's LLM brain. There should be a large mathetmatics model trained too, and integrated into the AI as a whole. This is the only way we will advance, not just with training larger and larger sets.
I completely agree, as you saw in my other post. But we don't need to train the llm with mathematical data. Just add an arithmetic and logic unit, like those in CPUs and create an interface that connects it to the LLM.
I wonder if I could do this using their API.
I'll try this idea.
The most important point of course is that the LLM is not the central agent of the Gestalt. But for now we will use the tools we have...