Back and forth with atebits over e-mail: >>I, personally, found the false positives much more acceptable than the >>current situation where you have to hunt for originating tweets for "false >>negatives".
>Doing anything interesting like automatically crawling conversation >webs is flat out impossible with false positives, and only an >annoyance with false negatives. It is a lie that it is "impossible" with false positives. With false positives, you can *always* crawl all conversation webs when they are correct, even when auto-linked, and you can easily tell when the auto- linking targeted an incorrect tweet. With false negatives, it's a lot worse because sometimes you can't crawl a correct conversation web at all. It is *far* faster for a user to identify a false positive then a user to hunt for a false negative. Again, it takes 1 second to identify that the auto-linking was incorrect, but 10 seconds to MINUTES to find the correct reply to a false negative, especially if the user is a prolific tweeter. Again, the new "in_reply_to_status_id" is relatively new, so with most people using that, the conversation webs will largely be correct. But when someone forgets to use the reply swoosh, I'd rather have Twitter auto-link the reply even if it causes some conversation webs to be falsely connected. >I would also argue that false negatives should be blamed on crappy >clients. I know that a few clients (up until recently) didn't set the >in_reply_to_status_id AT ALL, even for tweets where the user >*explicitly* replies to a particular tweet (i.e. clicked the reply >button next to it). I'm sorry, but also no. I have seen many people who are using conforming clients not jump through the UI hoops to perform a "correct" reply, both out of habit (i.e.: constant violators), or out of error (i.e.: just a one-time mistake). I prefer to take both kinds of human error out of the question via auto-linking. The false negatives were caused by people not used to the fact that they have to perform additional UI actions because of the change. To force users to do something to get a correct reply is stupid, in contrast to letting them do what they naturally do (which is how it was before). >There will be some growing pains, which will last as long as people >continue to use crappy clients. After that, many really interesting >things become possible. No, again, people are already using conforming clients. And, no, again, even with false positives, really interesting things are *already* possible. False positives do not inhibit any of those really interesting things. >That sounds rather hackish. I think the correct long term solution is >to leave it exactly as-is. The other thing I'd like to point out is >that with the old system, there was no way to express a "general" >reply. By that I mean a reply to someone that *isn't* in response to >a particular tweet... more of just a "directed" tweet - which is a >legitimately useful thing to express (and I'm not sure how you would >express it using your workaround). *facepalm* I am well aware that you couldn't express a general reply with the old system. Stop convincing yourself that I'm advocating to go back to the old way. With my way, you do it exactly as you do it now, and as you did it before: you simply type in "@atebits" and then your message. Twitter will auto-link it, and then display the link if the user's prefs say to display auto-links. The user can figure out whether the context is correct or not. The point is that humans are much more capable of determining whether context is correct or not, but computers are far better at establishing any sort of context in the first place. So the most effective way to establish the best context is to let both computers and humans do what they are best at doing. Computers will provide as much context as possible, and humans will throw out the context that isn't good. -- Simone
