If anyone's interested in a janky-but-works-great dictation setup on Linux, here's mine:

On key press, start recording microphone to /tmp/dictate.mp3:

  # Save up to 10 mins. Minimize buffering. Save pid
  ffmpeg -f pulse -i default -ar 16000 -ac 1 -t 600 -y -c:a libmp3lame -q:a 2 -flush_packets 1 -avioflags direct -loglevel quiet /tmp/dictate.mp3 &
  echo $! > /tmp/dictate.pid

On key release, stop recording, transcribe with whisper.cpp, trim whitespace and print to stdout:

  # Stop recording
  kill $(cat /tmp/dictate.pid)
  # Transcribe
  whisper-cli --language en --model $HOME/.local/share/whisper/ggml-large-v3-turbo-q8_0.bin --no-prints --no-timestamps /tmp/dictate.mp3 | tr -d '\n' | sed 's/^[[:space:]]*//;s/[[:space:]]*$//'

I keep these in a dictate.sh script and bind to press/release on a single key. A programmable keyboard helps here. I use https://git.sr.ht/%7Egeb/dotool to turn the transcription into keystrokes. I've also tried ydotool and wtype, but they seem to swallow keystrokes.

  bindsym XF86Launch5 exec dictate.sh start
  bindsym --release XF86Launch5 exec echo "type $(dictate.sh stop)" | dotoolc

This gives a very functional push-to-talk setup.

I'm very impressed with https://github.com/ggml-org/whisper.cpp. Transcription quality with large-v3-turbo-q8_0 is excellent IMO and a Vulkan build is very fast on my 6600XT. It takes about 1s for an average sentence to appear after I release the hotkey.

I'm keeping an eye on the NVidia models, hopefully they work on ggml soon too. E.g. https://github.com/ggml-org/whisper.cpp/issues/3118.

Ah yes—as the saying goes: “keep your friends at the Bayes-optimal distance corresponding to your level of confidence in their out-of-distribution behavior, and your enemies closer”

Like a caterpillar becoming a butterfly, an executive cannot graduate to an oligarch without failing upward into fabulous wealth. This is but another step on his beautiful journey.

Have you ever heard of "The Narcissist's Prayer"? It goes like this:

That didn't happen.

And if it did, it wasn't that bad.

And if it was, that's not a big deal.

And if it is, that's not my fault.

And if it was, I didn't mean it.

And if I did...

You deserved it.

Tether defenders are really working their way through the steps here.

18 months ago, it was "That didn't happen." (Tether is 100% backed by USD cash.)

6 months ago, it "wasn't that bad." (It might not be 100% USD cash, but it's cash-equivalent assets like short-term commercial paper.)

Now that there's strong evidence the commercial paper is just fake money shuffling between Tether/Binfinex/other shady crypto investments we get "that's not a big deal." (Look at the way banks work! They only need 4% collateral! Tether's probably got at least that much...)

Next step is finding out that their actual liquidity isn't capable of holding up under a real-life stress test, and the defenders will be talking about "not my fault." (This was a once-in-a-lifetime crash, they couldn't have foreseen it, crypto's still way better than the fiat banking system!)

When thousands of people lose their retirements in a gigantic defi crash, it'll be "you deserved it." (Everyone knows crypto is risky, you shouldn't have believed Tether was the same as USD.)