Explorer

RT @ericxtang: Kling releases a real time video AI model that uses long context memory to ensure consistency. Autoregressive video models…

open-source AI · protocols · programming · paragliding · building https://t.co/Cdhh8ECiyd · previously https://t.co/2fw43SSiR5 · 🇺🇸🏳️‍🌈

Conner Ruhl

Mon Dec 22 05:33:54

Harping on this point, but basically typing something usually requires knowing what you're going to say ahead of time, which isn't true for pacing around and brain-dumping through speech.

open-source AI · protocols · programming · paragliding · building https://t.co/Cdhh8ECiyd · previously https://t.co/2fw43SSiR5 · 🇺🇸🏳️‍🌈

Conner Ruhl

Sun Dec 21 23:42:19

Yeeting thoughts directly into models via voice is basically bootleg Neuralink. Give Superwhisper a try.

Harping on this point, but basically typing something usually requires knowing what you're going to say ahead of time, which isn't true for pacing around and brain-dumping through speech.

Conner Ruhl

Sun Dec 21 23:40:22

These are all non-trivial tasks. Coding has always been more of a means for me, not an end, so being able to create/use more code while writing less of it is amazing.

open-source AI · protocols · programming · paragliding · building https://t.co/Cdhh8ECiyd · previously https://t.co/2fw43SSiR5 · 🇺🇸🏳️‍🌈

Conner Ruhl

Sun Dec 21 23:35:16

I've long believed this kind of workflow would be possible, but it's still wild to experience that kind of shift because of a single model release. Several things hit >80% good enough all at once.

These are all non-trivial tasks. Coding has always been more of a means for me, not an end, so being able to create/use more code while writing less of it is amazing.

Conner Ruhl

Sun Dec 21 23:31:42

Programming via voice has always been a dream of mine, Cursor + SuperWhisper + Opus 4.5 has finally made it happen. I'm cycling through 3-5 active conversation threads and speaking more and more hours per day.

A super strict set of guidelines, domain-driven codebase, and Effect.ts means I can trust Claude to do huge amounts of work. I can use "domain speak" like "convert Assets.Preload to be a new Buffers domain which fetches and caches on the fly and use if for World.Characters."

Conner Ruhl

Sun Dec 21 23:21:35

Newest first — browse tweet threads

Explorer

Newest first — browse tweet threads

RT @ericxtang: Kling releases a real time video AI model that uses long context memory to ensure consistency. Autoregressive video models…

Harping on this point, but basically typing something usually requires knowing what you're going to say ahead of time, which isn't true for pacing around and brain-dumping through speech.

Yeeting thoughts directly into models via voice is basically bootleg Neuralink. Give Superwhisper a try.

These are all non-trivial tasks. Coding has always been more of a means for me, not an end, so being able to create/use more code while writing less of it is amazing.

I've long believed this kind of workflow would be possible, but it's still wild to experience that kind of shift because of a single model release. Several things hit >80% good enough all at once.

Programming via voice has always been a dream of mine, Cursor + SuperWhisper + Opus 4.5 has finally made it happen. I'm cycling through 3-5 active conversation threads and speaking more and more hours per day.