Mushroom foraging is a really fascinating area combining intuition with structure. Another fun source of inspiration: I am a member of a mushroom ID group on Facebook. Often, when a user posts a photo, it's not enough for an ID, so experts ask for specific information
They might ask, "where was this found?" Or "can you take a photo of the gills?"
A distinguishing feature of experts is the ability to quickly identify the right questions to ask to arrive at a positive ID more efficiently and with high certainty
Is there any work on this kind of thing for software systems? I'd like to use something like this to build proof automation at some point. But it also seems cool for lots of learning tasks
Interestingly, after observing the group for long enough, I have learned many of the right questions to ask when presented with an image that is not enough to ID a mushroom
So I play a game with myself when I see a mushroom photo: I try to identify it and, if I fail, I try to predict the questions that experts in the group will ask in the comments. Then I check and see if I was right
From this I've learned general principles like minimal pieces of information (say, what tree a mushroom is growing on) that distinguish edible mushrooms from poisonous lookalikes
This means I can ask other people those questions, but I can also ask myself those questions, so I've gotten really good at identifying edible mushrooms in the PNW
• • •
Missing some Tweet in this thread? You can try to
force a refresh
It's interesting by social class, some couldn’t find work, some people worked to support their families, some to support themselves with spending money, some just for the sake of learning work ethic, some for fun, and some were above working
Oh my gosh I wrote my generals report while I was so depressed I barely survived and honestly, the comments in the LaTeX are so snarky they're hilarious and I'm going to share some without context below
% like machine learning or whatever; but honestly, I doubt it
Why are people scared of PL? Especially type theory and proof assistants and so on? A professor recently said my work had a reputation of being "so hard basically nobody else can do it" and I'm flattered, but I also don't think it's true at all.
I just take Curry-Howard to its logical conclusion and treat proofs like programs. I diff them. I transform them. I compile proofs to new proofs. And it's actually really nice because proofs carry so much structure.
I do lack reasonable fear instincts about problems other people call hard. Maybe there's a social process involved in coming to view a problem as hard that I just ignore.
Even in vision I can think of examples that computers still are awful at that people are amazing at, and that people use algorithms for. Like ask any mushroom forager how we tell chanterelles apart from false chanterelles. There's a simple algorithm!
False gills, like those in chanterelles, fork and then merge again. True gills never do. You can avoid picking mushrooms that will hurt you (correctness matters a lot for mushroom foraging) if you check for just one fork and merge
Spotting things that look kind of like chanterelles is fuzzy and we do that pretty quickly and imperfectly. But once we do that the algorithm is what gives confirmation if there is doubt. Other keys: if there's one, there are many others nearby
It still confuses me that in spaces in which humans can often generalize from just 2 or 3 examples, it's considered successful when a software system does so from millions of examples
Nevermind robustness issues
It's true that, you know, reading the entire internet is something that computers can do better than us. But why should they have to? I feel like the metric for success is just wild
On the "we" versus "I" debate about my thesis, I ended up going with this:
- "I" for things I did,
- "we" for mathematical handholding, and
- "Nate" and "RanDair" and so on for things my coauthors did.
Nonstandard I guess, but I deliberately designed projects to decouple work.
So it is actually very easy to point to the parts that my coauthors did. And for the things that we really did design together, I will add a note about this in an early section at the end of the introduction, along with complete authorship statements.
I'm going to use the knowledge package that @16kbps recommended to introduce those authors and link back to the full authorship statements when I mention their names in later chapters. Credit is extremely important!!!