Rus

[Rus @tronpai]

Hey Bangkok, walking around the city is my favorite activity and I'm sure many of you like it too.

I want to preserve some routes and share it (many places around are changing so fast) as well as I want to learn other people favorite routes or places. 1/

[Rus @tronpai]

everything is changing...

[Karun Singh @karun1710]

We're hiring a Research Engineer at @Arsenal ⚽🔴⚪ to work directly with our Men's First Team!

We're building state-of-the-art AI models for the football domain. This role will focus on building the application layer for our research to advance coaching and analysis workflows.

[Rus @tronpai]

I need better hobbies

[Rus @tronpai]

@jxnlco Would be worry if it was Proust. Byung Han was conquered by the noon?

[Rus @tronpai]

@andimarafioti @cerebras Curious if you looking into speech to speech or it’s too early?

[Rus @tronpai]

have little idea what's happening, but it's very entertaining... "teaching a big language model to hear and eventually speak." keeps me pressing Y to everything it asks.

[Rus @tronpai]

First Midjourney Medical, now Figma with full codebase workflows...

[Rus @tronpai]

@kadirnardev have you thought to try with new gemma 4 12B?

[Rus @tronpai]

Thank you, now back into the terminal :)

I was digging around, found https://t.co/4DFXQALDwZ you might like to explore, same core thesis as your's Gemma approach (a single autoregressive LLM decides listen/speak/stop, no external VAD or turn-taking), but different design point.

[Rus @tronpai]

This is what people call a high-value reply! Thank you, it's concrete enough to validate several things I've been wrestling today.

One tiny engineering clarification: per 80 ms step, what conditions the depth transformer - Gemma's hidden state at the 2nd 40 ms audio token, a pool of the two 40 ms tokens, or a dedicated frame token?

[Rus @tronpai]

@matthen2 seems like @LorenzoIotti2 made it work

[Rus @tronpai]

Sleepless nights are back... teaching GPUs to generate speech sounds correctly.

Think of it like teaching someone to make the right mouth shapes and sounds - getting the pronunciation perfect.

What a time to be alive.

[Rus @tronpai]

@matthen2 You might like this conversation on this idea https://t.co/qjCR8ZJYLq

[Rus @tronpai]

@matthen2 That idea of audio in / out without the encoder occupied me this weekend. Good thing training isn’t that long/expensive + plus some labs opened their datasets as a bonus to ease tedious part.

[Rus @tronpai]

@LorenzoIotti2 came here with the same question! Was listening to founder(s) of Sesame and year ago they were saying that it's going that way and @sesame has a nice demo recently https://t.co/K2QclRkycP

[Rus @tronpai]

@WystanTBS This too shall pass. Always.

[Rus @tronpai]

Where can I buy an Argentinian IP address?

[Rus @tronpai]

haven't seen their actual numbers (similarweb).