Keynote: An OPEN Path to Super Intelligence
summary
An academic-led institute is building open foundation models that aim to match much larger closed models: K2 (32B parameters) launches with reported parity to models 20× larger and some closed models 100× larger, runs at 2,000 tokens/sec (≈10× faster), and fuels an Abu Dhabi hackathon with tens of thousands of developers. Development uses supervised fine-tuning, reinforcement learning with verifiable rewards, agentic planning, and test-time optimization, and promises an open-source generative latent predictive model world simulator and multimodal AIdriven digital organism for biology.
AI-written summary · about
chapters (15)
- 0:00OPEN super intelligence vision
- 1:51Institute of Foundation Models
- 3:14Redefining openness in AI
- 4:19K2 Think model breakthroughs
- 6:33Resource limits and clever techniques
- 7:15Future model releases
- 7:36Language model limitations
- 8:40World models for reasoning
- 10:50True world models beyond generative
- 11:32Generative latent predictive model
- 12:34Open source world model plans
- 12:55New agent model architectures
- 14:18AI for science and multimodal
- 15:45Drug design simulate first
- 16:27Aggressive open source AI strategies
related talks
03mKeynote: Open Models Need Open Memory
44mKeynote: Tracking the Pulse of Generative AI
44mKeynote: Tracking the Pulse of Generative AI
13mKeynote: Introductory Remarks
17mKeynote
25mState of Open Source AI Coding Models
15mKeynote: Building an Open Agentic Future
51mKeynote Panel: Open Innovation in AI - Present an...