I’ve just switched back from FUTO to GBoard after trying it out for the last month.
Both FUTO and GBoard use local ML inference to auto-correct words based on some fine tuned models. I do think FUTO auto-correct is pretty good, much better than the other FOSS keyboards that I’ve tried. From a technological standpoint they’re pretty tight.
I see that FUTO uses a fine tuned version of Meta’s open sourced LLaMA model to run on device which is indeed pretty great for this type of task. The unsolvable problem for FUTO and other FOSS keyboards is that they could never match the quantity of training data to better fine tune their models well enough to compete with the trillion dollar conglomerates like Google or M$ that have been gathering data from their keyboards for many years now.
GBoard has over 10B+ downloads on the Play store and will train on anyone’s keyboard usage data as long as they have the network enabled. That’s an absurd amount of data for ML fine tuning.
I do wish FUTO the best, but as a user, I want my keyboard auto-correct to work flawlessly. Given that there are no privacy concerns for me to just use GBoard with network turned off + no Google Play services, I see no reason to settle for an inferior product which in this case is unfortunately any other FOSS keyboard.