The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Big Science and Embodied Learning at Hugging Face đ¤ with Thomas Wolf - #564
4 snips
Mar 21, 2022 Thomas Wolf, co-founder and chief science officer at Hugging Face, shares his fascinating journey from quantum physics and patent law to machine learning. He discusses the BigScience project, which unites over 1,000 researchers to create a vast multilingual dataset, emphasizing the importance of diverse data and ethical AI. The conversation dives into the innovations of transformers in NLP, multimodality, and the implications for the metaverse. Thomas also touches on his new book and the evolving landscape of AI, advocating for collaborative and responsible advancements.
AI Snips
Chapters
Books
Transcript
Episode notes
Hugging Face Origin Story
- Thomas Wolf transitioned from quantum physics to patent law before co-founding Hugging Face.
- Initially a chatbot gaming company, Hugging Face pivoted to open-source, driven by the success of tools like Deepmoji.
Hugging Face Expansion
- Hugging Face, initially focused on NLP, has expanded into speech and computer vision.
- This expansion is driven by the limitations of text-only models and the convergence of tools like Transformers.
BigScience Project Origins
- The BigScience project, spurred by an underutilized compute cluster in France, aims to build a large, public language model.
- Inspired by large physics collaborations like the LHC, it involves over 1,000 researchers and a substantial compute grant.
