I’ve been having this hunch lately that using LVMs (GPT-4V, LLaVa, etc) could be a solution to error correcting, assuming the LVM has a sufficiently good enough

I’ve been having this hunch lately that using LVMs (GPT-4V, LLaVa, etc) could be a solution to error correcting, assuming the LVM has a sufficiently good enough world model