Also, open weight models follow the same pattern: can't be stopped easily but small chunk of people will have what it takes to use them, and without approved hardware you are crippled. Tbf this one is...
Jan3 recently made a phone app for speech to text. The app is very easy to use and works really well, even on a fairly old phone: https://github.com/jan3dev/a1echos.
Handy.computer (speech to text) works amazingly well too, even on an old desktop.
ppq.ai also made it easy to use end-to-end encrypted models and pay privately with Lightning, without an account. I have personally put many friends on ppq.ai, even friends who really don’t consider themselves technical.
trymaple.ai is also an interesting option.
I think the LLM memory wiki trend is also very inspiring. Done well, it seems likely to allow the use of less powerful models and still get pretty good results.
Handy.computer (speech to text) works amazingly well too, even on an old desktop.
ppq.ai also made it easy to use end-to-end encrypted models and pay privately with Lightning, without an account. I have personally put many friends on ppq.ai, even friends who really don’t consider themselves technical.
trymaple.ai is also an interesting option.
I think the LLM memory wiki trend is also very inspiring. Done well, it seems likely to allow the use of less powerful models and still get pretty good results.