Yeah, so you are agreeing that the benchmarks are useless because they don't answer those questions.

Can AI models generalize+ at any long context problem solving and agency regardless of modality? I think the answer is no, and this is why they are not yet AGI.

+ generalize being the key word.