I saw him, briefly, at the Media Lab's 30th anniversary celebration at the end of October. He was clearly in frailer condition than when I'd seen him last, but still you don't expect a fixture like that to be silenced forever.
I took a class from Marvin, who did not like his students calling him Professor Minsky, even the ones he did not house, mentor, and shepherd personally. If you were a professional colleague then "Dr Minsky" was alright but we learned to call him Marvin. I forget what the class was called but it really doesn't matter, because Marvin didn't so much "teach" a "class" as stand in front of a packed hall for three hours one night a week and do performance art.
It was almost all verbal - he never much bothered with slides or visuals except a diagram here or there - and it all appeared to be extemporaneous. He'd just... start talking and keeping up with him was unpredictably easy or difficult depending on how "on" he was that night. His topics ranged from theories of mind to the history of artificial intelligence to science fiction to current trends in computer science to why Americans are so bad at teaching children simple math. You couldn't predict what he was going to riff on for any given night, and that was just fine.
He was also an unabashed curmudgeon whom you'd find once in a while playing snippets of various symphonic pieces on the piano that used to be in the Lab atrium. I think it helped him organize his thoughts, though his music rambled as much as his lectures - he'd start things in mid-musical phrase and just trail off or stop if something else caught his fancy. Talking to him had to wait until he did stop, unless you were a pretty senior colleague or one of his favored students. He was hard to deal with oftentimes: abrasive and blunt, and stubbornly set in his ways.
My favorite memories of our few 1:1 conversations revolve around books. I'd tell him I had read another new book and he'd challenge me: "Tell me two new ideas that are in that book." I'd try and he'd dismiss them with "Nonsense" or "That can't be right." That didn't mean he actually thought the idea was nonsense so much as he thought I hadn't done a good job summarizing or framing the novelty of the idea. Minsky did love science fiction, with the strong emphasis on the 'science' part. He once averred that science fiction was about ideas and other fiction was about the seven deadly sins. Whether a book was per se labeled as science fiction didn't matter to him so much as whether it contained interesting ideas - once he'd determined that it did he'd apply his own label.
And finally, I remember and treasure the one assignment I did for his class that he called out for praise. We were discussing why humans have such a strong leaning toward comparing only two things, and why we go to such great lengths to simplify things into two comparables. What, he asked, is wrong with comparing three things?
I realized (and wrote in my answer) that if you are comparing A and B you only have one mental comparison to handle, A-B. Whereas if you compare three things, you have to manage three mental comparisons: A-B, A-C, and B-C. This, I pointed out, meant that pairwise comparisons involved an information reduction and thus the brain ought to favor them over three-way comparisons that did not reduce information load. He told me my answer was "interesting" and "might even be right". High praise indeed from Marvin.