Generating AI videos for enterprise with Synthesia CEO Victor Riparbelli | StrictlyVC London

()

Generative AI Discussion

Ingrid London and Victor Riparbelli had a fireside chat about generative AI at a recent conference.
They discussed the controversy surrounding OpenAI's use of Scarlett Johansson's voice without her permission.
Synthesia takes consent seriously and does not clone anyone's voice or visual likeness without their explicit consent.

OpenAI's growth strategies include creating impressive demo videos and having no content moderation or safety controls, which can help them go viral.
There are concerns about the potential negative impact of OpenAI's actions on companies like Synthesia, which focuses on enterprise AI video platforms for internal videos and customer service.

Synthesia believes that different AI companies will focus on different areas, with OpenAI aiming to be an infrastructure layer and Synthesia providing applied AI solutions for specific problems.
While foundational models are getting a lot of attention, Synthesia emphasizes the importance of building great products that solve real problems for customers.
Synthesia has a research team working on deep AI research, but their focus is on a narrow slice of the market: human presenters in videos in corporate contexts.
Synthesia believes that customers care more about specific features and improvements that solve their problems rather than the underlying technology and parameters of the AI models.
The focus should be on the customer and their needs, not on competitors like OpenAI or Anthropic.

Synthesia is open to collaborating with B2B companies, including social media platforms like TikTok and Snapchat, to integrate AI-generated content into their products.
The company offers an API for developers to incorporate avatars into their applications.
Synthesia believes that successful video evolution will involve a shift from static mp4 files to more dynamic and interactive video formats.
The company is currently focused on improving the quality of storytelling capabilities through avatars.
High-profile people may eventually be added to the platform, but the technology is not yet good enough for widespread adoption.
The company has a strong business model and is not急于筹集资金.

Building a moat in the AI industry is challenging due to the rapid advancements of AI models like LLMs and diffusion models.
The speaker believes that building a foundational model should be done in the traditional manner, focusing on creating a product that is an integral part of the customer's infrastructure.
Open-source models like SAR V are not controllable and are built for a different purpose, such as creative generation rather than consistency or specific instructions.
The speaker's team has scanned thousands of people in a 3D studio over the past two years, giving them a unique dataset that provides an advantage in creating videos.
The speaker emphasizes that these technologies should not be considered the end goal or the main product, but rather a temporary advantage in terms of speed to market.