I think there are constrained sub-problems that are more solvable. One that looks particularly tractable is strict subsetting, in which the text remains linear but some readers want to read a subset (e.g. without the math). We have decent interface solutions there.
Conversation
The more significant problem with that very constrained framing or the somewhat less constrained multiple-threads framing is in authoring. Quality prose dances; arcs can’t readily be truncated or transposed. It appears to be very hard to compose well in these mediums.
2
6
Possibly there’s a chicken-and-egg problem. We’ve developed extensive methods, partly tacit, for structuring linear text—developed collectively over centuries, and individually through reading examples and gradually getting better at producing them.
3
3
There's the fundamental constraint that our attention is limited ant time is linear, so you are always reading linear text; even when meandering in meaningness.com, one is still reading a sequence of texts. It seems it's inescapable
1
3
to have to linearize our thoughts. One can provide itineraries or recommended sequences to venture through the domain, or else it's up to the reader to navigate a graph which can be daunting
1
3
What I see as working better is linear plus varying width, so you can read the same but deeper, make each step of the journey simpler or more complex depending on how much you want to invest. (i.e. post X has 3 progressively more complex variants)
2
Yes… but maybe making an explicit conceptual dependency graph (“you can only read this web page if you already understand X, Y, and Z; follow the links for those if you don’t”) is worth a try.
1
1
Oh, cool, thank you!
Is there any sort of community consensus on how well this worked out?
1
Badly! But idk why exactly, the view was that it failed relative to its goals but idk what the goals exactly were
1
They had a very interesting postmortem, as it happens:
Amusingly, I disagree with the founder! He clashed w Eliezer, who wanted to focus on specific content, do special-case stuff, perfect that. Alexei thinks they should have focused on getting more authors. Premature in my view: prove in the high-effort case, then generalize.
1
3
It’s sad: in the end, we’re left with no clear insight about what the best-case problems were with that core idea of surfacing and navigating dependency chains.
2
3
Show replies