lunixbochs's profile picture. My opinions are the views of your employer.

Ryan Hileman

@lunixbochs

My opinions are the views of your employer.

مثبتة

"say hello" to the Talon 0.4 release: ported app from Qt/C++ to Rust, all-new Conformer D speech model, better eye tracking, gamepad support, more! Update post: patreon.com/posts/talon-0-… Release notes: talonvoice.com/dl/latest/chan… Support Talon on Patreon! patreon.com/lunixbochs/mem…

This video is not sped up. Updating demo to show off better "Conformer D" model coming in Talon 0.4. This video took 48s, i.e. 36% faster. First demo had careful pauses on "dollar brace", no longer needed. Also see how fast I say "push bang space close tag third lap" at 00:08



user: dictation takes 45 seconds to insert two sentences into win11 notepad. verdict: this is a regression in notepad. classic notepad doesn't have this problem. also the problem stops if you rapidly wiggle your mouse over the window.


git is shaming me for my habit of taking notes by just running `vim` and then letting the vim process die later when I reboot

lunixbochs's tweet image. git is shaming me for my habit of taking notes by just running `vim` and then letting the vim process die later when I reboot

Ryan Hileman أعاد

I wrote about all my favourite cool tracing methods, including eBPF, QEMU’s JIT, Tracy, magic-trace and how to easily visualize your own trace data in Perfetto! Also includes some tools I wrote and an example eBPF use at Anthropic. thume.ca/2023/12/02/tra…


can diffusion generate model weights?


Unscientific M1 Max vs M3 Max efficiency core benchmark on 14" MacBook. Launched Geekbench 6 with `taskpolicy -c background`. browser.geekbench.com/v6/cpu/3490272 browser.geekbench.com/v6/cpu/3489709

lunixbochs's tweet image. Unscientific M1 Max vs M3 Max efficiency core benchmark on 14" MacBook. Launched Geekbench 6 with `taskpolicy -c background`.

browser.geekbench.com/v6/cpu/3490272
browser.geekbench.com/v6/cpu/3489709

Ryan Hileman أعاد

cursedCTF 2024 Halloween Teaser 31/10/23 - 01/11/23 this CTF teaser committed tax fraud against multiple 501c5 organizations


Playing with adding Stream Deck support to Talon


How will LLMs change the way we interact with computers in 5 years? (Input % is amount of time you spend typing/clicking on computer, vs time spent asking LM to perform tasks and then waiting for it). Reply with your relevant background.


Ryan Hileman أعاد

"Speech @Wikimedia: A 77 Language Multilingual Speech Dataset": 1780 hours of CC-BY-SA licensed transcribed speech from a diverse set of scenarios and speakers, in 77 different languages. (Mosquera et al, 2023) paper: dmlr.ai/assets/accepte… data: huggingface.co/datasets/MLCom…

WikiResearch's tweet image. "Speech @Wikimedia: A 77 Language Multilingual Speech Dataset": 1780 hours of CC-BY-SA licensed transcribed speech from a diverse set of scenarios and speakers, in 77 different languages.

(Mosquera et al, 2023)

paper: dmlr.ai/assets/accepte…
data: huggingface.co/datasets/MLCom…

This video is not sped up. Updating demo to show off better "Conformer D" model coming in Talon 0.4. This video took 48s, i.e. 36% faster. First demo had careful pauses on "dollar brace", no longer needed. Also see how fast I say "push bang space close tag third lap" at 00:08

Conformer better handles accents as well as fast speech. Here's a demo dictating Vue code at high speed with the new model, with no errors. Compared to two typists on the same code: 80wpm typist took 1m54s, 120wpm took 53s. It took me 1m15s with voice. I think I could go faster.



Loading...

Something went wrong.


Something went wrong.