Episode description

Video: https://youtu.be/dVIPBxHJ1kQ

00:00 Intro

00:15 Greets for Sonam

01:02 Importance of metric learning

3:37 Sonam's background: Rasa, Qdrant

4:31 What's EmbedAnything

5:52 What a user gets

8:48 Do I need to know Rust?

10:18 Call-out to the community

10:35 Multimodality

12:32 How to evaluate quality of LLM-based systems

16:38 QA for multimodal use cases

18:17 Place for a human in the LLM craze

19:00 Use cases for EmbedAnything

20:54 Closing theme (a longer one - enjoy!)

Show notes:

- GitHub: https://github.com/StarlightSearch/EmbedAnything

- HuggingFace Candle: https://github.com/huggingface/candle

- Sonam's talk on Berlin Buzzwords 2024: https://www.youtube.com/watch?v=YfR3kuSo-XQ

- Removing GIL from Python: https://peps.python.org/pep-0703

- Blind pairs in CLIP: https://arxiv.org/abs/2401.06209

- Dark matter of intelligence: https://ai.meta.com/blog/self-supervised-learning-the-dark-matter-of-intelligence/

- Rasa chatbots: https://github.com/RasaHQ/rasa

- Prometheus: https://github.com/prometheus-eval/prometheus-eval

- Dino: https://github.com/facebookresearch/dino