Yeah, they definitely should focus on adding more RAM to their flagship phones, but smaller models are ideal for smartphones.

Smaller models are quicker to run and use up less battery. Besides, if they’re using some AI models I would rather it be run locally than have my data uploaded to some server somewhere (big assumption that they wouldn’t do that anyway, I know)… Or at the very least I can still run it even if the network goes down.

I had the same problem, but I found that turning up the advanced Parameter Values in the Text Prediction settings helps to keep everything spelled correctly. I’m currently at 10.4 for the transformer strength and 8.3 for the autocorrect threshold and I’m noticing a lot less issues… I haven’t tested out turning it up all the way yet.

I also have the Transformer fine tuning enabled.

The swipe feature definitely needs some attention.

They do have a crowdsourced way of helping to improve their swipe prediction, more info on the link here: https://swipe.futo.org/

You’ll notice that even the swipe prediction for Heliboard comes from a closed source library.

Futo has been my favorite so far compared to other keyboards. I haven’t used Heliboard yet, but it also looks like a good option.