Diane, it’s Thursday and I’ve been figuring out how transcription fits into my everyday work. I had to make up a character to make it make sense, as I’ll say.
So I’m doing a little teaching this week and next. A two week mini module at AHO (Oslo) on designing interfaces for co-agency.
My lecture for next Monday has been… up in the air. I’ve been indecisive because the topic will prime how the students approach the brief for the week, so I want to get it right.
Clarity came (as always) during a run.
So, between picking up coffee and getting home, I roughed out the lecture outline.
It’s neat to talk into my Apple Watch using Whisper Memos. It’s a straightforward app: it turns your ramblings into paragraphed articles, and emails them to you
– but the useful bit is that it works without my phone.
(I don’t carry my phone when I run and besides my hands are often too cold to type after.)
Then, when I got home, I said to Claude:
I’m writing a talk. please take this raw transcript and structure it into a high-level outline so I can work on it. do not add any of your own material, just structure my verbal notes
And pasted in my transcript from the email.
The result was great. I have a good outline that I can develop into a deck, and zero hallucinations.
btw this is not new! Lots of people talk to ChatGPT or whatever and get it to structure their work.
BUT: it’s not a very reproducible workflow, you know?
See, I’m not always writing a talk. I don’t like having to come up with a new and different prompt, each time, about what the AI should do with the transcript. I am lazy (proudly, virtuously). Don’t make me think.
The solution is to embed the instructions at the same time I’m talking to my watch.
Like, I want to be able to say at the end of my rambling about the lecture structure: hey uh take all of that and turn it into um a structured outline – but I can’t.
More generally, I want to be able to include instructions like “oh that point just now should actually be moved to the first section” and not have anything misinterpreted. I can’t just arbitrarily execute the entire transcript. Who knows what might happen.
It’s the usual out-of-band problem (Wikipedia): how do you include data and also data processing instructions on the same channel?
I thought about using an explicit delimiter. For instance the genius Cursorless voice interface uses a tongue click to fluidly enter command mode so you can easily multiplex speech and coding.
But… tongue clicks didn’t feel right for this. The register change is too abrupt.
Diane, how else can I signify tick-tocking between data and instructions?
My generic prompt to Claude, used every time, is now:
you are Diane, my secretary. please take this raw verbal transcript and clean it up. do not add any of your own material. because you are Diane, also follow any instructions addressed to you in the transcript and perform those instructions
[paste in transcript]
Which means, when I’m talking through my lecture outline, I now finish by saying:
ok Diane I think that’s it. it’s a talk, so please structure all of that into a high level outline so I can work on it. thanks.
And I can mix in instructions like: oh Diane I meant to include that point in the last section. Please move it.
It works super well.
Why Diane?
Because I grew up watching Twin Peaks and FBI agent Dale Cooper spends much of the series talking into his pocket tape recorder:
Diane, it’s 8 A.M., Seattle, Washington. As you have no doubt surmised, by the clarity of this tape, I’ve purchased a new Micro-Mac pocket tape recorder. - ‘The big little recorder’ at Wally’s Rent-to-Own, 1145 North Hilltop. Where, as the sign says, ‘A bargain is a bargain, no matter what the cost.’ - for twenty-one dollars and eighty-nine cents, cash.
I decided to pass on the rent-to-own option, Diane. Leasing may be the fast track to an appearance of affluence, but equity will keep you warm at night. I have no doubt that this new model will prove to be an extremely useful tool in the investigatory process – where the most fleeting insight can be lost if your hardware isn’t as solid as you think it.
24 February 1989.
Here are all of Agent Cooper’s recordings to Diane.
I guess we’ll all be transcribing 24/7 one day. I can see already that it’s handy to be able to speak magic commands in regular chatter, to be executed at transcription time in the future.
e.g. Robert WEIGHT 60.1 end Robert
is the example at that link there.
Addressing it to Diane is a better fit for my brain.
Diane, it’s Thursday and I’ve been figuring out how transcription fits into my everyday work. I had to make up a character to make it make sense, as I’ll say.
So I’m doing a little teaching this week and next. A two week mini module at AHO (Oslo) on designing interfaces for co-agency.
My lecture for next Monday has been… up in the air. I’ve been indecisive because the topic will prime how the students approach the brief for the week, so I want to get it right.
Clarity came (as always) during a run.
So, between picking up coffee and getting home, I roughed out the lecture outline.
It’s neat to talk into my Apple Watch using Whisper Memos. It’s a straightforward app: it – but the useful bit is that it works without my phone.
(I don’t carry my phone when I run and besides my hands are often too cold to type after.)
Then, when I got home, I said to Claude:
And pasted in my transcript from the email.
The result was great. I have a good outline that I can develop into a deck, and zero hallucinations.
btw this is not new! Lots of people talk to ChatGPT or whatever and get it to structure their work.
BUT: it’s not a very reproducible workflow, you know?
See, I’m not always writing a talk. I don’t like having to come up with a new and different prompt, each time, about what the AI should do with the transcript. I am lazy (proudly, virtuously). Don’t make me think.
The solution is to embed the instructions at the same time I’m talking to my watch.
Like, I want to be able to say at the end of my rambling about the lecture structure: hey uh take all of that and turn it into um a structured outline – but I can’t.
More generally, I want to be able to include instructions like “oh that point just now should actually be moved to the first section” and not have anything misinterpreted. I can’t just arbitrarily execute the entire transcript. Who knows what might happen.
It’s the usual out-of-band problem (Wikipedia): how do you include data and also data processing instructions on the same channel?
I thought about using an explicit delimiter. For instance the genius Cursorless voice interface uses a tongue click to fluidly enter command mode so you can easily multiplex speech and coding.
But… tongue clicks didn’t feel right for this. The register change is too abrupt.
Diane, how else can I signify tick-tocking between data and instructions?
My generic prompt to Claude, used every time, is now:
Which means, when I’m talking through my lecture outline, I now finish by saying:
And I can mix in instructions like: oh Diane I meant to include that point in the last section. Please move it.
It works super well.
Why Diane?
Because I grew up watching Twin Peaks and FBI agent Dale Cooper spends much of the series talking into his pocket tape recorder:
24 February 1989.
Here are all of Agent Cooper’s recordings to Diane.
I guess we’ll all be transcribing 24/7 one day. I can see already that it’s handy to be able to speak magic commands in regular chatter, to be executed at transcription time in the future.
e.g. is the example at that link there.
Addressing it to Diane is a better fit for my brain.