←
dadabots8,848 posts
dadabots avatar

dadabots

@dadabots

∿ Music hackers. Algoraves. Inventing, playing. Neural synthesis. 24/7 ai death metal. Stable Audio team. Open models. Mischief @harmonai_org @artblocks_io

◷ Joined October 2017▧ 1,081 Photos and videos
dadabots avatar
dadabots@dadabots·May 20

🥳 Announcing Stable Audio 3 🍕 🏆 fastest music models ever 💻 runs on MacBookPro M-series 🧪 break it plz 🧠 LoRA finetune in < 1h 📷 Sm = faster, Medium = qualityer ⚡ 59x realtime on M5 Pro One-liner fast install: curl -LsSf dadabots.com/_/sa3-mac | bash

0:17

Meet Stable Audio 3.0, the open-weight model family built for artistic experimentation. This is our open invitation to experiment with generative audio. We believe the best innovations are still waiting to be built. The 4-1-1 on 3.0: 📣 You own your outputs, and can distribute and commercialize them under the Stability AI Community License (up to $1 million in revenue). 🎵 New and improved capabilities include variable-length generation up to six minutes, and full song composition on portable devices, no GPU required. ✅ Trained on a fully licensed dataset. 🎨 You can customize the models on your own library with support for LoRa training, which we’ve documented for the first time. More on the models 👇

0:10
dadabots avatar
dadabots@dadabots·11h

Imagine music AI like this

PyBot: No weights. No training. A chatbot built with only simple rules and no AI, entirely written in 600k lines of Python code. PyBot matches Smollm with 360M parameters! How? A swarm of 1,000 DeepSeek Flash agents wrote the whole code in just a week. github.com/huggingface/pybot

1:42
erin ➡️ sf nov28-dec10 avatar
⟳ dadabots retweeted
erin ➡️ sf nov28-dec10@erinbeess·Nov 17

various other experiments from the last few days (have not had time to sit down and really get big progress done)

Post attachmentPost attachment

masked decoding on gemma 3 12b to control text rivers if a token isn't the length i want it to be, i mask it out added a lora for width tags so the model is aware of text length + coherence RL afterwards

Post attachment
dadabots avatar
dadabots@dadabots·12h

Listen to everyone's tracks & spread the link around if you could plzzz 🥹👉👈

dadabots avatar
dadabots@dadabots·Nov 13

“….The best thing I've heard machine intelligence produce is still @dadabots excellent Calculating Calculating Infinity, released all the way back in 2017 with SampleRNN which is an ancient model.…”

With audio models (or in this case just... opus 5.5 with a guitar), it seems like every newer, larger model is a step backwards. This particular qt'd example is just not something anyone would ever listen to, and I think part of it is that these models just end up producing something that sounds like nothing at all because of how large and generalized they are. Every new Lyria/Suno model seems to be trained in a way where the outputs are heavily weighted towards some generalized "average preference". So the fidelity increases, you can be more specific, and somehow it all still sounds completely uncompelling. The best thing I've heard machine intelligence produce is still @dadabots excellent "Calculating Calculating Infinity", released all the way back in 2017 with SampleRNN which is an ancient model. Some of it is genuinely novel, retaining a distinct sound signature... and it's actually "hard" lol (check out this breakdown on Variations on a Cocktail Dressfinity from all the way to the end of the song, fucking nuts tbh) Not saying that AI music won't ever be good, but I have a feeling that "good" AI music will come from a retvrn to smaller models trained on very specific source material.

0:46
dadabots avatar
dadabots@dadabots·Nov 10

hm how would you write the algorithm

just realized it is now technically possible to throw a party where every guest has a litemotif and the music in each room dynamically weaves everyone's themes together

Jameson Hodge avatar
⟳ dadabots retweeted
Jameson Hodge@jamesonhodge·Nov 7

my vocal synth models the vocal tract as a network of branching acoustic tubes, unlike vocaloid which stitches together and pitch-shifts phoneme recordings. still working on improving naturalness by tuning it against recordings of real singers, but I'm already impressed at how cool it sounds!

2:25

I only make serious software

0:39
sagnik avatar
⟳ dadabots retweeted
sagnik@datavorous_·Oct 31

I manually inspected 40,000 animations to post train an LLM to make demoscenes. Scraped 10 years of demos, did continued pretraining, curriculum learning and finally RLHF-ed with myself as the reward function. Yes, models can have "taste". All you need is data. 1/6

Post attachment
lyra avatar
⟳ dadabots retweeted
lyra@_lyraaaa_·Oct 31

go read findings dot md paragraph 9 immediately wtf this is amazing

a9lim@_a9limOct 30

@treetowntree and i created neuron lens: a database of every single neuron and attention head in qwen 3 8b and gemma 4 12b read using the R-lens (a J-lens variant) and what we call the "S-lens". you can find it at neuron-lens.com/, along with the tools we used to make it. using the R-lens to read neurons' input and output projections turns out to be very effective at determining what a neuron does. we're able to legibly read what a majority of neurons do when fired: the lens qualitatively predicts the result of steering over 90% of the time! the same trick lets us read attention heads' QK circuits and OV circuits, showing us both what they attend to and what they do with it. however, the R-lens has two important flaws: its accuracy degrades in early layers, and it can't separate the MLP block and the attention block. we found ways around both these issues. we first modified the r-lens by fitting it on input tokens rather than output tokens to get the s-lens, which is accurate in early layers but inaccurate in late layers. we then fitted R- and S-lenses to the activations after attention but before the MLP to get "mid-lenses", which are much better at deciphering MLP inputs and attention outputs. using both of these techniques, we were able to interpret most of the models' layers.

0:23
Sauers avatar
⟳ dadabots retweeted
Sauers@Sauers_·Oct 29

Introducing the neural twins: models trained to be made for each other, that communicate entirely in neuralese. 🧵

1:49