Just in case anyone here was wondering how Reddit’s numbers are looking these days…
Data and visuals from https://subredditstats.com/r/askreddit
Just in case anyone here was wondering how Reddit’s numbers are looking these days…
Data and visuals from https://subredditstats.com/r/askreddit
Hello fellow Gboard user. I would have also noticed “abs” in place of “and”.
Fucking thing gets worse every goddamned day. Thanks, lol
It does feel this way… but why?! Shouldn’t it be getting better with more training data? Is other people’s shitty typing data fucking up my experience?
Or maybe my swiping technique is getting lazy? I’m not sure, but it definitely feels worse than a few years ago.
If you make the same typo often enough and don’t correct it with the autocorrect feature (eg backspacing or jumping to the wrong character instead) Gboard will think it’s what you’re meaning to type and add it to your dictionary. When it comes up in the little autocorrect thing above your keyboard, you can drag it into the trash. You could also go through your dictionary and delete specific words as well. That’s what works for me anyways, not sure if I’m correct
Wait what? I can delete suggestions‽
Holy shit, I just tried it! WTF‽‽
Happy to help! Bonus tip, on Android you can hold the space bar down and slide the cursor left or right to bring it to specific letters too. On apple you can jump lines with the cursor by doing the same thing, just up/down instead
Yeah, I’ve been using the spacebar cursor thing for years, only after a similar interaction. It gets annoying at times since it will often speed up way past the point I’m trying to go, however.
It could be that the people’s typing (and therefore the training data) includes so many different unique movements that an “average” of all of them doesn’t actually look like any real person’s individual typing pattern. Sorta how some one-size-fits-all designs don’t fit any one person perfectly.
This seems plausible to me.
They should make it smarter and group users based on their swipe technique and have a separate model for each group.