I sometimes find myself agreeing (or disagreeing) with only part of a long comment.
Current options include replying with a quotation of the part in question, or voting on the whole thing based on the part. Sometimes this isn’t worth the effort or seems unfair and I wish I could just vote on that one part.
Conciseness in comments is a virtue given our limited free time and attention, but I don’t want to turn Less Wrong into Twitter. Due to inferential gaps, some concepts really do take long comments to get across, but aren’t worth a top-level post. Our karma system should accommodate that.
I don’t think it should be so fine-grained that we could insert a vote between every character. That’s noise. It could really make a popular post hard to read. On the fine scale, per-sentence, or perhaps even per-punctuation mark is more reasonable.
But I suspect that the appropriate granularity is per paragraph. There’s already a visual gap there, so you can insert karma votes (or tags) without hurting readability.
I’m not sure how to handle the case of edits. Minor typo fixes aren’t a problem, but changing the number of paragraphs, or moving sentences around could be a problem. But there’s nothing stopping us now from completely rewriting a comment to say something completely different. This is bad form, of course. And it might be bad form to change that in a deceptive way once it’s voted on, but honest mistakes are possible.
I think edited posts (even comments) should have a publicly-visible history, like a wiki.
I don’t know if this is worth the cost of implementing, but I absolutely like the idea. In particular, in the hypothetical world where this was in place, if one had e.g. two karma to spend per post (upvote strength of two) they could at most upvote or downvote two particular paragraphs, such that they’d left the same “weight” of opinion, but in a more detailed way.
Oliver, Ben and I were actually talking about this a few days ago. (I actually ran into the “man I really wish I could upvote this paragraph and downvote this other paragraph a week ago)
A few thoughts came up (epistemic status: current thoughts, nothing clearcut)
a) it might incentivize certain writing styles. Exact details depend on the execution, but regardless you’re fiddling with Goodheart’s Demon. It could have weirder consequences than you intend. And this might subtly make people with different writing styles (in particular ones with more meandering styles rather than clearly bulleted points) feel like the site didn’t want them to write the way they felt most comfortable, which leads to them writing less.
b) you might be able to get most of the same value by implementing claims as an object type. i.e. let people write their article without worry about how pieces of it will get upvoted/downvoted, but afterwards you have the option of adding a summary of explicit claims, which get listed as short bullet points and potentially top-level-comments to respond to. Those claims can be either upvoted/downvoted, or potentially (see Robby Bensinger comment somewhere on this page), get a “probability this is correct” a la Arbital.
The Claim Thing is also fiddling with incentives that might have unintended consequences, but seemed like it might be pointing in a direction that was a) easier to implement, and b) if done well could have other valuable downstream effects, c) you probably don’t want to implement both because they are occupying similar niches, and each feature adds to the cognitive load of the site.
Per-Paragraph Voting
I sometimes find myself agreeing (or disagreeing) with only part of a long comment.
Current options include replying with a quotation of the part in question, or voting on the whole thing based on the part. Sometimes this isn’t worth the effort or seems unfair and I wish I could just vote on that one part.
Conciseness in comments is a virtue given our limited free time and attention, but I don’t want to turn Less Wrong into Twitter. Due to inferential gaps, some concepts really do take long comments to get across, but aren’t worth a top-level post. Our karma system should accommodate that.
I don’t think it should be so fine-grained that we could insert a vote between every character. That’s noise. It could really make a popular post hard to read. On the fine scale, per-sentence, or perhaps even per-punctuation mark is more reasonable.
But I suspect that the appropriate granularity is per paragraph. There’s already a visual gap there, so you can insert karma votes (or tags) without hurting readability.
I’m not sure how to handle the case of edits. Minor typo fixes aren’t a problem, but changing the number of paragraphs, or moving sentences around could be a problem. But there’s nothing stopping us now from completely rewriting a comment to say something completely different. This is bad form, of course. And it might be bad form to change that in a deceptive way once it’s voted on, but honest mistakes are possible.
I think edited posts (even comments) should have a publicly-visible history, like a wiki.
I don’t know if this is worth the cost of implementing, but I absolutely like the idea. In particular, in the hypothetical world where this was in place, if one had e.g. two karma to spend per post (upvote strength of two) they could at most upvote or downvote two particular paragraphs, such that they’d left the same “weight” of opinion, but in a more detailed way.
Oliver, Ben and I were actually talking about this a few days ago. (I actually ran into the “man I really wish I could upvote this paragraph and downvote this other paragraph a week ago)
A few thoughts came up (epistemic status: current thoughts, nothing clearcut)
a) it might incentivize certain writing styles. Exact details depend on the execution, but regardless you’re fiddling with Goodheart’s Demon. It could have weirder consequences than you intend. And this might subtly make people with different writing styles (in particular ones with more meandering styles rather than clearly bulleted points) feel like the site didn’t want them to write the way they felt most comfortable, which leads to them writing less.
b) you might be able to get most of the same value by implementing claims as an object type. i.e. let people write their article without worry about how pieces of it will get upvoted/downvoted, but afterwards you have the option of adding a summary of explicit claims, which get listed as short bullet points and potentially top-level-comments to respond to. Those claims can be either upvoted/downvoted, or potentially (see Robby Bensinger comment somewhere on this page), get a “probability this is correct” a la Arbital.
The Claim Thing is also fiddling with incentives that might have unintended consequences, but seemed like it might be pointing in a direction that was a) easier to implement, and b) if done well could have other valuable downstream effects, c) you probably don’t want to implement both because they are occupying similar niches, and each feature adds to the cognitive load of the site.