Comment by fsiefken | Hacker News Reader

Comment by fsiefken | Hacker News Reader

Parent Full thread

Colemak was just released when I decided to switch to a better keyboard layout to prevent a return to RSI. I looked into rolling my own based on statistics. I didn't care if it would look like QWERTY, if I needed to switch cold turkey for my health or future comfort I'd do so. I decided to use the tried and true DVORAK and tweak a minimum to make it even beter; switch the U (qwerty F) and the I (qwerty G) as the I is used much more frequently in both English and Dutch. Next to being optimized for finger travel the DVORAK layout has more so called 'hand alteration' during typing due to the vowels being on the left.

It took months before I was 'up to speed'. I thought I would type significantly faster, but that is'nt the case as I often think slower then that I can type. It's much more about comfort. I don't regret it, it really is more comfortable. I switch between QWERTY and DVORAK-IU often so I am bilingual. Typing in alphabetic shorthand (for example Yash) with yet another optimized layout is another possibility. Or Yash with morse code or braille mapped like chording or just chording. https://www.artofchording.com/layout/chorded-keyboard.html

Soon we will be talking to our LLM's a lot, so it's not a real big deal, but the nerd in me likes to optimize. Still I think, perhaps I could have designed the very best layout for my personal style and Dutch en English and the latest theories regarding what's more optimal (rolls, hand alteration, compression, chording). Perhaps an LLM can help? Not sure if I want to learn a new layout again. In the meantime alternating between QWERTY and DVORAK-IU suits me well.

> Soon we will be talking to our LLM's a lot, so it's not a real big deal

Regardless of how good voice interfaces get, the inherent limitations and restrictions that come with it means that we'll all still be doing a whole lot of typing until/unless a new and better input method is invented.

The only replacement for typing I can see taking over would be a mind interface, but even there I have a suspicion that it'll require 'thinking in a different voice' that'll make it almost as tiring as voice control.

I had the same thought. Maybe it would be like our internal monologue voice (which I learned not everyone has). So instead of just one in there, we'd have two.

Or maybe it would be a combination of internal voice and ephemeral voiceless conception, or maybe thinking with "intent". Whatever it would be, it would certainly take effort to some degree.

I don't know about anyone else, obviously, but my internal monologue is very poorly focused at times, I can be mentally editing code and suddenly think about donuts, or squirrels.

Maybe AI could filter that nonsense out based on contextual clues, but not the AI we have so far, hence I think the monologue would need that second voice. 'Remember Nikki, think all code things and only code things in scottish... Aye, I shall.... och, bollocks!'

sublingual vocalisation to the rescue - i think there is work being done on translating that to something able to be used to control stuff :)

SRV https://atelier.net/ve-tech-radar/tech-radar/subvocal-recogn...

https://web.stanford.edu/class/cs224s/project/reports_2017/P...

https://annals-csis.org/Volume_11/drp/pdf/153.pdf

https://en.wikipedia.org/wiki/Subvocal_recognition

I don't really like voice interfaces, but I see that as a personal preference.

What sort of limitations/restrictions do you see with it, beyond accuracy (which is continually improving)?

Accuracy incurs secondary problems today; in my experience, I type slower than I can talk, but I type much faster than I can talk to a computer and have any hope of being understood. Now, I do have high-ish hopes that this problem will be rendered obsolete in the next few years, but it remains a problem so far. The other thing that occurs to me is sort of connected and also something that might be fixed soon; I can type arbitrary symbols and words that aren't in the dictionary with no extra overhead, and I don't trust a machine to handle that. However, I suspect that sufficiently accurate voice-to-text combined with even current LLMs may be up to the task of handling that, so this whole comment really might boil down to "give it 5 years and we'll see".

Accuracy isn't one of them, because that's not something inherent to using voice as an interface. I'm thinking of aspects that are inherent to communicating by voice at all, not even related to machines understanding it.

Such as... voice is really awful at communicating with great precision and exactness. It can certainly be used for that, but it rapidly becomes tedious and very time-consuming. Using a keyboard is faster, more accurate, and more convenient for that sort of thing.

> Soon we will be talking to our LLM's a lot

People have been saying this for decades. I think of "dragon naturally speaking" and similar. It still doesn't seem mainstream. (maybe it is but not visible to me?)

that said, I still remember this video:

https://youtu.be/YKuRkGkf5HU

I simply don't like a speech-based interface, but I think it's unfair to compare STT tech from even a few years ago to something like Whisper. It's dramatically improved in both accuracy and speed.

It's, inadvertently, aptly named. Soon, all of us in the office will be literally whispering to our headsets all day. Can't wait! /s

On the bright side, it will settle the debate on tabs vs. spaces once and for all.

Reply on news.ycombinator.com