Key Concepts
- Generative AI models (Veo, Flow)
- Storytelling and emotional content
- Collaboration between artists and technologists
- Primordial Soup (Darren Aronofsky's company)
- AI-native filmmakers
- World models and AI understanding of physics
- Hallucination vs. Imagination in AI
- Move 37 (AlphaGo) and AI creativity
- Democratization of filmmaking
- LoRAs (Low-Rank Adaptation) for style transfer
- Hybrid approach to filmmaking (live-action + AI)
- Interactive storytelling and gaming (Holodeck)
Primordial Soup and the Google DeepMind Partnership
Darren Aronofsky discussed his strategic partnership with Google DeepMind through his company, Primordial Soup. He was inspired by the potential of generative models to impact storytelling and wanted to ensure storytellers were at the forefront of using these new tools. He chose to work with Google DeepMind because of Demis Hassabis's focus on human benefit. Aronofsky believes storytelling is a fundamental human art and wants to empower storytellers to create new narratives.
The name "Primordial Soup" reflects the emerging and unstable nature of the technology. Their motto is "Make soup, not slop," emphasizing the need to shape the raw output of AI into emotionally resonant stories. Aronofsky is working on three films using these technologies, collaborating with both veteran and AI-native filmmakers.
The Importance of Collaboration and Human Creativity
Demis Hassabis emphasized the importance of collaboration between technologists and artists. He believes that creative input is essential for shaping AI tools in a positive direction and giving them "soul." He highlighted the need to understand what artists want from these tools, down to the technical level, and to work together to achieve that.
Hassabis also noted that technological development is inherently creative and that there aren't hard lines between the two fields. He drew a parallel to the Renaissance era and figures like da Vinci, who seamlessly integrated art and technology. He believes that creativity is what separates good scientists and artists from great ones.
Veo 3 and the Future of AI
Hassabis discussed the development of Veo 3, Google DeepMind's generative video model. He explained that the goal is to build tools that empower the creative process. He emphasized that the creativity and spontaneity come from the user, who can achieve different results from the same tool through skill and mastery.
From an AGI perspective, Hassabis highlighted the importance of "world models," where AI understands the physical environment and the physics of the world. He believes that generating accurate physics and light is a key step towards this goal. He also addressed the concept of "hallucination" in AI, noting that it can be useful in creative contexts as a form of imagination.
Move 37 and AI Creativity
Hassabis discussed the significance of Move 37 from AlphaGo, which he sees as an example of AI generating a truly novel idea in a heavily studied area. However, he also acknowledged that current AI systems are still far from achieving true out-of-the-box thinking and invention. He posed the question of whether AI could have come up with general relativity or invented the game of Go, concluding that it couldn't.
He mentioned AlphaEvolve as an example of a system that uses evolutionary techniques to go beyond what current foundation models know. He believes that hybrid systems, combining different AI techniques, will be important for future exploration.
The Democratization of Filmmaking and the Role of Storytelling
Aronofsky and Hassabis discussed the potential for AI to democratize filmmaking, giving more opportunities to storytellers around the world. However, Aronofsky emphasized that storytelling still requires craft, patience, and collaboration. He believes that it will never be just a person with prompts on a computer.
Aronofsky also highlighted the importance of storytelling in a world where attention is increasingly fragmented. He argued that storytelling connects us and allows us to relate to people from different backgrounds and experiences. He expressed concern about the consequences of losing our attention to meaningful stories.
Eliza McNitt's "Ancestra"
Eliza McNitt discussed her film "Ancestra," which is being created in collaboration with Primordial Soup and Google DeepMind. The film tells the story of the day she was born and nearly died. She described it as a personal story about the ferocity of a mother's love.
McNitt emphasized the importance of capturing human performances and emotions in her film. She used a hybrid approach, combining live-action footage with AI-generated visuals. She also used AI to generate scenes with a baby, as it would be unethical to work with a real baby on set.
McNitt trained a style using her late father's photographs to incorporate his vision into the film. She expressed excitement about the potential for AI to help people connect with their families in new ways.
LoRAs and Style Transfer
Hassabis explained how LoRAs (Low-Rank Adaptation) can be used to bring in specific styles and content into the AI generation process. He praised McNitt's use of her childhood pictures to create a personalized style for her film.
The Future of Storytelling and Interactive Experiences
Looking to the future, Aronofsky and Hassabis speculated about the new forms of storytelling that might emerge with AI. Aronofsky believes that AI will enable very personal stories to be told. McNitt hopes that AI will democratize filmmaking, making it accessible to more people.
Hassabis expressed excitement about the potential for interactive storytelling and gaming experiences, drawing a parallel to the Holodeck from "Star Trek." He believes that combining AI-generated visuals with interactive elements will create new possibilities for entertainment and storytelling.
Conclusion
The conversation highlighted the transformative potential of AI in filmmaking and storytelling. The speakers emphasized the importance of collaboration between artists and technologists, the need to shape AI tools in a positive direction, and the potential for AI to democratize filmmaking and enable new forms of creative expression. They also acknowledged the challenges and ethical considerations that come with these new technologies. The collaboration between Darren Aronofsky, Demis Hassabis, and Eliza McNitt serves as a model for how AI can be used to empower storytellers and create meaningful experiences.
AI summaries can miss context or contain errors. Check important details against the original video.