3 ms·
Once you start letting anyone edit, you have to solve a bunch of other problems: various kinds of spam, non-spam low quality edits, deconflicting manual edits w
by jdm2212 6y ago
Once you start letting anyone edit, you have to solve a bunch of other problems: various kinds of spam, non-spam low quality edits, deconflicting manual edits with other data sources. The easy/default solution for most things is to have a trustworthy sources only (official website, Google My Business page, etc.) and a decision tree for choosing which one wins.
The Maps feature for editing listings was a ton of work spanning a bunch of teams over several months, but that kind of investment is made only after every other option has been exhausted because it's so slow and expensive and scales poorly relative to just ingesting more data and throwing ML at it.
- luckylion 6y agoSure, giving everybody direct editing access isn't a good idea. I'm just surprised that so many somewhat user-interaction features feel sub-standard. The maps or gmail gui are super polished, so obviously there are people who do that kind of thing, but others feel rushed, like they were given to an intern that really wanted to leave for the weekend. E.g. the Knowledge Boxes on SERPs. When I was invited to add/edit to them for a notable client a few months back, it was barely documented, cumbersome, error and response messages were vague and it felt like a proof of concept. I guess I'm just wondering: is that by design, has someone tested that and figured out "if people feel like they're using ebay's dysfunctional interface, they'll spend more money on ads", or is it just that somebody gets a task "hey, add a form to edit that, please" and they're like "well, they didn't ask me to make it pretty and I really need to leave soon"?
- jdm2212 6y agoOh yeah I'd bet crappy UIs are just laziness/sloppiness/prioritization rather than malice. The glamorous frontend work that gets lots of design attention is the stuff that has billions of users. I saw some A/B testing for ad engagement that drove UI decision making, but it was always short-term and related to search -- like, "we updated the model today, and users click on the ad 1% more often in the treatment group". FWIW I was pretty much at the bottom of the totem pole so if there was some nefarious plot to provide crappy UIs I wouldn't've been in the loop anyway.