teropa
banner
teropa.bsky.social
teropa
@teropa.bsky.social
510 followers 1.8K following 220 posts
One half of Counterpoint. Interests: software, music, mind.
Posts Media Videos Starter Packs
Laznia Nowa Theatre, Unsound Krakow
Reposted by teropa
The KickStarter is live!
www.kickstarter.com/projects/190...
Our big fat beautiful book, Designing Imaginative Environments is a deep dive into the world of
environment design for storytelling.
Reposted by teropa
I've been re-acquainting myself with NNs, transformers and LLMs this week, so these are early thoughts. But I am starting to think about language without the capacity to imagine language, i.e., an LLM unable to imagine itself participating in language. mail.cyberneticforests.com/what-machine...
What Machines Don't Know
Imagining Language Without Imagination It's important to acknowledge that Large Language Models are complex. There's an oversimplified binary in online chatter between the dismissive characterizatio...
mail.cyberneticforests.com
10s latent sequences, mean-pooled to one 64-dimensional vector each, then UMAP-reduced to 3D. Neighbour-hopping in the UMAP space.
I computed Stable Audio Open latent space embeddings for my Bandcamp music collection.

A nearest-neighbour traversal reveals fun connections.
Jim O’Rourke & Eiko Ishibashi
Reposted by teropa
Brian Eno and Ezra Klein discuss the impact @stevereich.bsky.social's music has had on them on the latest episode of @nytopinion.nytimes.com's 'The Ezra Klein Show.' www.nytimes.com/2025/10/03/o...
Visualising principal components of Stable Audio Open latents over time.

0.8s hann window, 0.03s hop (latent rate ≈21.5 Hz)
Reposted by teropa
Today’s episode of The Ezra Klein Show.

The musician and record producer Brian Eno delves into his experiments with ambient music, his thoughts on generative A.I. and his deep gratitude for the uniqueness of human life.
open.spotify.com/episode/1Dio...

youtu.be/AfYY9v0Q0X4?...
Brian Eno on How Machines Can Make Us More Human
YouTube video by The Ezra Klein Show
youtu.be
still feel compelled to say "i might be wrong" when i'm pointing out to it that things look fishy. this is to try and stave off the sycophancy, where it just wants to agree with me over getting it right.
sonnet 4.5 seems to respond quite well to explicit "give me some chain-of-thought on this" requests. self-corrected nicely with this slightly delicate bit of logic.
sonifying the wind on a great circle journey around the globe
putting sonnet 4.5 through its paces as an ambient melody composer
Reposted by teropa
Generative features are utilized to enhance audio understanding; audio generation models encode both spatiotemporal perception and semantic prior; a fusion strategy is proposed; experiments demonstrate consistent performance gains in sound event classification, tagging, and audio captioning.
When Audio Generators Become Good Listeners: Generative Features for Understanding Tasks
Zeyu Xie, Chenxing Li, Xuenan Xu, Mengyue Wu, Wenfu Wang, Ruibo Fu, Meng Yu, Dong Yu, Yuexian Zou
arxiv.org
Reposted by teropa
this is the really ineffective head-in-the-sand reaction a lot of folks on Bluesky tend to have if you talk about any harm reduction-based approach to the reality that millions of people are using AI tools today. The current approach to critiquing AI is obviously not working, but they don't care.
Reposted by teropa
HP Spaceball 2003 (1991). While the mouse operates on X and Y axes, the HP Spaceball also used the Z axis, making it easy to rotate objects in 3D space. Plus the white plastic made it look like something out of 2001: A Space Odyssey.