Post AaGTLKVG31rSmDRL96 by joeldrapper@ruby.social
(DIR) More posts by joeldrapper@ruby.social
(DIR) Post #AaGSwlB3jjlSGXHCiW by simon@fedi.simonwillison.net
2023-09-29T16:15:24Z
0 likes, 0 repeats
I'm on the latest episode of the Rooftop Ruby podcast with @collin and @joeldrapper talking about Large Language ModelsIt was a really excellent conversation - we covered a huge amount of groundI'm trying something new: I put together my own transcript with Whisper, then cleaned that up and added inline links and section headings. Here's the result, complete with an embedded audio player that can jump to each different section: https://simonwillison.net/2023/Sep/29/llms-podcast/
(DIR) Post #AaGTLKVG31rSmDRL96 by joeldrapper@ruby.social
2023-09-29T16:20:40Z
0 likes, 0 repeats
@simon @collin it was so great you could join us. Thank you! The transcript looks great. How did you do speaker diarization?
(DIR) Post #AaGTXOLtYyowBmoQOu by simon@fedi.simonwillison.net
2023-09-29T16:21:09Z
0 likes, 0 repeats
@collin @joeldrapper I used GPT-4 to help build my own custom audio player, with a 3x speed button!Here's that GPT-4 transcript: https://chat.openai.com/share/4ea13846-6292-4412-97e5-57400279c6c7
(DIR) Post #AaGTtAcvH4aTzXA5Mu by simon@fedi.simonwillison.net
2023-09-29T16:26:56Z
0 likes, 0 repeats
@collin @joeldrapper Here's the full list of topics we covered. You can click through to each of these to jump directly to that point in the audio (or just read the annotated transcript) https://simonwillison.net/2023/Sep/29/llms-podcast/
(DIR) Post #AaGVPnoTDStQkKgePY by simon@fedi.simonwillison.net
2023-09-29T16:44:09Z
0 likes, 0 repeats
@joeldrapper @collin Meticulously by hand in MacWhisper - it took ages! Really looking forward to finding a good automation mechanism for that.
(DIR) Post #AaGWd6XuyitIZ3qrKa by simon@fedi.simonwillison.net
2023-09-29T16:57:44Z
0 likes, 0 repeats
@collin @joeldrapper From the podcast, here are my thoughts on whether leaning on LLM assistance is likely to help or hurt new programmers:https://simonwillison.net/2023/Sep/29/llms-podcast/#does-it-help-or-hurt-new-programmers
(DIR) Post #AaGXCudKkMJA2jSWzA by joeldrapper@ruby.social
2023-09-29T17:04:03Z
0 likes, 0 repeats
@simon @collin ah, I see. It looks like MacWhisper can do it if you have separate audio tracks for each speaker. We could probably have got you these. 😅
(DIR) Post #AaGXNdFvhDibKicjKa by mattmay@mstdn.social
2023-09-29T17:05:04Z
0 likes, 0 repeats
@simon With the side benefit of not having to deal with all the asshole seniors and StackOverflow jockeys who can’t wait to mock you for needing help
(DIR) Post #AaGXkVN3B9xtXQrRIW by simon@fedi.simonwillison.net
2023-09-29T17:10:11Z
0 likes, 0 repeats
@mattmay hard to overstate how important that it - being able to ask dumb questions with zero chance of judgement is wonderful
(DIR) Post #AaGZKrg1p4TF6AdhUe by joeldrapper@ruby.social
2023-09-29T17:26:10Z
0 likes, 0 repeats
@22 @simon 100% agree. The quality of transcription and annotated links etc. is exceptional. Now I want to make every @rooftop episode like this.
(DIR) Post #AaGZKsitvsMYLMzVbs by simon@fedi.simonwillison.net
2023-09-29T17:27:50Z
0 likes, 0 repeats
@joeldrapper @22 @rooftop I was really hoping I could get Claude or GPT-4 to do most of the work for me, but I couldn't find a prompt that worked - GPT-4 started inventing things people hadn't actually said!
(DIR) Post #AaGZnObF0gkZJb0M88 by joeldrapper@ruby.social
2023-09-29T17:32:53Z
0 likes, 0 repeats
@simon @22 @rooftop I bet it would write a pretty decent summary though. Or help you pull out the most interesting quotes.
(DIR) Post #AaGa0IxDJNfjZPRCRk by osma@sigmoid.social
2023-09-29T17:33:15Z
0 likes, 0 repeats
@simon@collin @joeldrapper Once again you've done an an amazing job, both in terms of content and form! I really liked the transcript but I would never had listened to such a long podcast. Thanks so much!
(DIR) Post #AaGaC0bLN9YS7omz9U by simon@fedi.simonwillison.net
2023-09-29T17:36:35Z
0 likes, 0 repeats
@joeldrapper @22 @rooftop I tried a bit of that too: "illustrate each section with a quote" - Claude did OK at that but not enough that I ended up using it
(DIR) Post #AaGhRcqQAP2wl0EdoO by parkr@fosstodon.org
2023-09-29T18:58:29Z
0 likes, 0 repeats
@simon @collin @joeldrapper I wish every podcast did this!!! Timestamps / chapter markers are the absolute best.
(DIR) Post #AaGjJqOa6ISMNOd9rU by baclace@sigmoid.social
2023-09-29T19:19:40Z
0 likes, 0 repeats
@simon @collin @joeldrapper I find LLM code gen works best as a way to generate standalone functions with unit tests that can be integrated into existing apps. As such, it is like generating a custom library instantly.Generating whole apps is much harder, especially if a UI is involved. It can be quite tedious to explain all the UI behaviors of a webapp at this point. Ultimately, eventually it will boil down to specifications-as-programming which has long been a goal.
(DIR) Post #AaGkD6Ujwf3IEbrS6q by simon@fedi.simonwillison.net
2023-09-29T19:29:23Z
0 likes, 0 repeats
@baclace @collin @joeldrapper Yeah, UI programming is definitely a lot less well served - I can get bits and pieces out of it, but fundamentally these models don't have a great idea of 2D space yet so they're not the best for interface workI'm looking forward to seeing if that changes with the new GPT-4 image inputs
(DIR) Post #AaGliJavOGmPnICUZk by jeppe@uddannelse.social
2023-09-29T19:46:32Z
0 likes, 0 repeats
@simon @collin @joeldrapper really great show. I learned a lot. Thank you for sharing.
(DIR) Post #AaGmHi4YNG7HqWw8MS by anirvan@mastodon.social
2023-09-29T19:53:00Z
0 likes, 0 repeats
@simon That transcript of using GPT4 to build the page was fascinating! I’d never thought of using it to do anything design oriented. Thanks for sharing your process.
(DIR) Post #AaGwb7gAOONeJ8QRJg by tomayac@toot.cafe
2023-09-29T21:48:23Z
0 likes, 0 repeats
@simon You don't say "please"? 😱
(DIR) Post #AaGx59ZTvDOurbcOPI by simon@fedi.simonwillison.net
2023-09-29T21:53:52Z
0 likes, 0 repeats
@tomayac I've tried to argue in the past that it's immoral to say "please" or "thank you" to a chatbot because it's anthropomorphizing them in a dangerous wayThe counter-argument to that (which I quite like) is that it's good to use them to practice being polite for when you're talking to actual humans
(DIR) Post #AaGxNcCME4F6rSbapU by tomayac@toot.cafe
2023-09-29T21:57:23Z
0 likes, 0 repeats
@simon I was joking of course, but to your point, Google Home devices, at least for a while, would positively note it to kids when they asked nicely. (I still always say “please” to machines. You never know. M3gan and friends are just around the corner.)
(DIR) Post #AaHQjlpE0lEZlBhaHQ by rj@arvr.social
2023-09-30T03:25:50Z
0 likes, 0 repeats
@simon The style of replies in your transcripts is quite distinctive, do you mind sharing your custom instructions?
(DIR) Post #AaHRZdKapxceTU6X1k by simon@fedi.simonwillison.net
2023-09-30T03:35:44Z
0 likes, 0 repeats
@rj I don't use any custom instructions - might be because I'm using Code Interpreter / Advanced Data Analysis
(DIR) Post #AaIGVGjU0lQGtGQGQa by ryanfb@digipres.club
2023-09-30T13:06:14Z
0 likes, 0 repeats
@simon FWIW, here's the completely unmodified output of whisperX with diarization for 3 speakers (medium.en for transcription). Takes ages on CPU but completely hands-off, should be much faster on GPU: https://gist.github.com/ryanfb/e602f1c52ca8bd0090d7b9ce456ba3de
(DIR) Post #AaIQ05ulBLGPu3TFwm by simon@fedi.simonwillison.net
2023-09-30T14:51:55Z
0 likes, 0 repeats
@ryanfb OK that's really impressive! Is there a good CLI tool for running that?
(DIR) Post #AaIQAq8GCTmw8muRSC by simon@fedi.simonwillison.net
2023-09-30T14:53:04Z
0 likes, 0 repeats
@ryanfb OK this looks usable, will try this soon https://github.com/m-bain/whisperX#english