OpenAI o3, DeepSeek-V3, and the Brundage/Marcus AI bet

The video discusses whether deep learning is experiencing stagnation, featuring opinions from industry professionals. Chris Hay expresses strong skepticism, stating models are worsening. Kush Varshney acknowledges a challenge but believes progress is ongoing, while Kate Soule argues that innovative applications of deep learning are emerging. They also delve into OpenAI’s new model, o3, discussing its implications for AI performance standards and governance.

**Key Points:**

Chris Hay claims deep learning models are getting worse, indicating the technology has “hit a wall.”

Kush Varshney sees a challenge but believes significant progress is being made through innovation.

Kate Soule highlights new applications of deep learning that will provide benefits by 2025.

The panel discusses OpenAI’s o3 model and its potential impact on AI performance benchmarks.

Inferences from the o3 model indicate that improvements are moving towards handling tasks more efficiently.

Concerns about AI governance arise, especially with the faster iteration of models and the emergence of open-source alternatives like DeepSeek-V3.

Experts believe that innovation in AI governance will require global cooperation and technical expertise, not just national laws.

The discussion of bets on AI capabilities showcases the rising expectations and differing perspectives on what AI can achieve.

There is skepticism about AI’s ability to meet high creative standards such as winning a Pulitzer Prize.

Chris asserts that the focus should shift from model prestige to the tasks they perform efficiently.