You are touching on the subject of attention, which is actually extremely important and has a whole field of research (see for example, the hillarious invisible

You are touching on the subject of attention, which is actually extremely important and has a whole field of research (see for example, the hillarious invisible monkey experiment: http://www.theinvisiblegorilla.com/gorilla_experiment.html http://www.theinvisiblegorilla.com/gorilla_experiment.html) There are a lot of things that we ignore or don't register consciously. It is possible to decode items in one's working memory in controlled situations (see for example: https://pubmed.ncbi.nlm.nih.gov/28155212/ https://pubmed.ncbi.nlm.nih.gov/28155212/). So I suppose one possible approach would be to hold things in your working memory and have the Neuralink actively decode those. But in that case, as you've pointed out, requires a lot of attention and most likely would rely more heavily on contextual AI to narrow the decoder search space. The basic premise of my skepticism with the "we need higher bandwidth I/O" statement still stays though -- the bottleneck of human communication is how quickly coherent thoughts can be formed in a way someone else can understand it (output), and how quickly information can be understood (input). For the latter example, you can scan an entire page in less than second. With immense speed reading practice, you can get a "gist" of a page in around that time. But understanding the details in that time is not possible.