Seems like LLMs are very good at generalizing to random tasks that they're not necessarily trained for. I wonder how GPT-4 would do at responding to inputs from LIDAR sensors on an autonomous vehicle.
We already have to deal with Tesla FSD hallucinations and regulating it in serious and safety critical applications such as autonomous transportation.
I don't think we need a LLM behind the wheel and hallucinating directions to end up leading the driver to drive off a cliff or running past a stop sign it got confused over.
> Seems like LLMs are very good at generalizing to random tasks that they're not necessarily trained for.
Some tools are useful for other applications, especially safety critical applications. LLMs are *absolutely* not useful for this use-case.
We already have to deal with Tesla FSD hallucinations and regulating it in serious and safety critical applications such as autonomous transportation.
I don't think we need a LLM behind the wheel and hallucinating directions to end up leading the driver to drive off a cliff or running past a stop sign it got confused over.
> Seems like LLMs are very good at generalizing to random tasks that they're not necessarily trained for.
Some tools are useful for other applications, especially safety critical applications. LLMs are *absolutely* not useful for this use-case.