Uploaded September 2026 | Updated November 2026, 3 days ago
Frontier AI models score 80 to 90 percent on standard benchmarks like MMMU, yet fall to pieces on visual tasks any 6-year-old handles effortlessly, like counting objects in an image or navigating a maze.
At Ray Summit 2026, Andrew Dai, co-founder of Elorian AI, draws on 12 years at Google Brain and DeepMind, where he co-led development on GLaM, PaLM 2, and Gemini, to describe the challenges in visual reasoning for current models and a new approach to solving them.
You'll leave understanding why the gap exists and what closing it takes.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLZNaJyYZRPdo
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
đź”— Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com
Frontier AI models score 80 to 90 percent on standard benchmarks like MMMU, yet fall to pieces on visual tasks any 6-year-old handles effortlessly, like counting objects in an image or navigating a maze.
At Ray Summit 2026, Andrew Dai, co-founder of Elorian AI, draws on 12 years at Google Brain and DeepMind, where he co-led development on GLaM, PaLM 2, and Gemini, to describe the challenges in visual reasoning for current models and a new approach to solving them.
You'll leave understanding why the gap exists and what closing it takes.
Liked this video? Check out other Ray Summit breakout session recordings: youtube.com/playlist?list=PLZNaJyYZRPdo
Subscribe to our YouTube channel to stay up-to-date on the future of AI: youtube.com/c/anyscale
đź”— Connect with us:
LinkedIn: linkedin.com/company/joinanyscale
X: https://x.com/anyscalecompute
Website: anyscale.com

