Figure’s AI Robot, AI News, Cognition’s Devin, and the biggest AI bet yet! | E1915

Data Transparency

The discussion on training data transparency highlights the complexities and legal implications of using various data sets for AI models. David Friedberg and Jason Calacanis critique OpenAI's handling of questions about their training data, particularly the response from their CTO regarding Sora's data sources. Jason questions the CTO's vague answers, suggesting a lack of media training and transparency, which could lead to legal challenges 1. David argues that the organization, not just the individual, bears responsibility for these communication failures, especially given OpenAI's ongoing legal battles with major media outlets 2.

You can't wing it when you're OpenAI and in a lawsuit with the New York Times.

--- Jason Calacanis

The conversation underscores the need for clear communication strategies in AI companies to navigate the legal landscape effectively.

   

Model Capabilities

Exploring advanced AI model capabilities reveals their potential to process complex inputs and make decisions. David Friedberg explains how large language models (LLMs) can perform reasoning tasks by accessing a selection of tools, enabling them to execute actions like identifying and handing over an apple from a table 3. This integration of physical agility and cognitive processing marks a significant advancement in AI's practical applications.

These LLMs now have something called function calling capabilities.

--- David Friedberg

Additionally, Jason Calacanis discusses how AI can learn from diverse sources, such as watching TV, to enhance its functional understanding and decision-making abilities 4. This approach broadens the scope of AI learning beyond traditional methods, paving the way for more intuitive and versatile AI systems.