rgabrielsson

3 ms·

1. 1.▲
   by rgabrielsson 1y ago

   This is my TEDxMIT talk on Large Behavioral Models, trained on actions rather than text. I know I’m pushing the narrative a bit hard, lol... but I’d genuinely love to hear any questions or feedback
2. 2.▲
   Actions speak louder than LLMs: A case for behavioral AI [video](youtube.com)

   4 points by rgabrielsson 1y ago|1 comments
3. 3.▲
   by rgabrielsson 1y ago

   Great question! It's three things: 1) Data. Every next event can be one out of billions of potential actions, so our vocabulary are in billions instead of in 100K as for LLMs 2) Architecture. Every action is multimodal, it consist of i
4. 4.▲
   by rgabrielsson 1y ago

   Transformer-based model with 150 M params, trained on 15 B purchases, and showed 10× lift in conversion,
5. 5.▲
   by rgabrielsson 1y ago

   15 B purchases, 10× lift, 150 M params
6. 6.▲
   A Foundation Model for Consumption, Transactions and Actions(research.unboxai.com)

   7 points by rgabrielsson 1y ago|5 comments
7. 7.▲
   by rgabrielsson 4y ago

   One trend I'm seeing: I think the math that is becoming more useful, if huge foundation models are taking over, is around comparing and understanding their embedding/latent spaces. Geometry, topology, optimal transport, alignment
8. 8.▲
   by rgabrielsson 4y ago

   haha future will tell! :D
9. 9.▲
   Self-Supervised Learning and Foundation Models(futureofai.mit.edu)

   35 points by rgabrielsson 4y ago|5 comments