Very very little of the math in the universe has been solved
Infinitesimally small
We have literally no idea how reality works
It’s time to get started
Kokoro TTS, but 10× smaller 🫰
Paradee distills Kokoro-82M into a 8M-param TTS
just 9 MB of weights can speak faster than real time on a single CPU thread and runs on your toaster
▶️ on Spaces hf.co/spaces/hugging…
Moats are misunderstood
It’s not about keeping competitors out
It’s about keeping users in
What makes your thing so sticky that they come and can never leave?
It’s a question about network effect and preventing churn
Embeddings are like a general vibe check. They turn text into number blobs that show overall similarity, but exact words can throw them off.
Decision models are like a scorecard. They rate specific things you pick (skills, interests) with clear scores and reasons.
Example:
Here’s an experiment you might get some value from
Embedding models are trained on a large corpus of data and each element is an abstract representation
Decision models can output a list of concrete representations, so you can generate vectors which are specific to the
