From: allin
The emergence of generative AI has initiated a significant shift in content creation, with implications for various industries and a rise in intellectual property issues. Companies like Google and Microsoft are integrating generative AI into their core products, fundamentally altering how information is accessed and monetized [00:33:18].
AI Capabilities in Content Creation
Google has developed “incredible AI competency,” particularly since acquiring DeepMind, primarily focusing on internal optimizations like data center energy efficiency, ad optimization, and YouTube algorithms [00:35:01]. This capability was largely geared towards preventing disruption in their primary search business [00:35:43].
OpenAI’s ChatGPT offers an alternative to traditional “information retrieval” search, which involves scanning, indexing, and ranking static data [00:35:57]. While traditional search presents lists and structured data (like maps or shopping results), ChatGPT’s natural language responsiveness is “fantastic and better” for specific answer solutions [00:38:16].
New generative AI tools are transforming various aspects of content and software development:
- Text and Code Generation: ChatGPT can generate text and code, even perfect Excel formulas, with high proficiency [01:06:06]. GitHub Copilot, a Microsoft acquisition, assists engineers by filling in code as they write it, significantly increasing productivity [01:11:51].
- User Interface (UI) Design: Tools like Galileo AI allow users to describe a UI design (e.g., “an onboarding screen of a dog walking app”) and generate a visual output [01:11:19].
- Media Production:
- Music: AI can write lyrics in the style of any artist and recreate voices, allowing for mashups like David Guetta playing an Eminem-voiced track [01:15:42].
- Video Games: The entire video game industry could be “rewritten with AI,” shifting from publishers making single games to tools allowing individuals to create and consume their own games [01:14:58].
- Movies: The cost of making a movie could drop significantly due to AI, potentially from $10 million to “ten thousand or a thousand dollars of compute time” [01:17:25]. This could democratize content creation, allowing individuals to create films in their basements [01:17:28].
- Automated Assistants: AI is expected to lead to “little personal digital assistants” embedded in every application, often voice-based, to help users accomplish tasks [01:06:46].
This impact of AI on software and services allows for “hyper efficiency” and “margin destruction” in middleman businesses, as human capital costs are reduced, allowing more capital to be pushed into customer acquisition [01:09:00].
Economic and Business Model Implications
The economic viability of generative AI in search is a significant concern due to cost:
- Google’s traditional search costs approximately 2.5 cents per query [00:39:13].
- Running a GPT-3 model for ChatGPT costs about 30 cents per result, an “order of magnitude higher” [00:39:28].
- To scale ChatGPT across all search queries, it would cost an estimated $80 billion a quarter from a compute perspective today [00:40:35]. This cost needs to decrease by 10x to be economically competitive [00:39:44].
This cost reduction is expected to naturally occur due to advancements in specialized silicon (massively parallelized computing) and cheaper energy costs [00:41:30].
AI as a computing platform and its potential disruption to search revenue models:
- Google’s business model relies on users clicking on paid links in search results [00:48:47].
- If AI provides direct answers (e.g., “top three televisions”), users may not need to click on ads or links, thereby “degrading” Google’s business quality [00:49:55].
- Google’s stock dropped significantly after its Bard demo, reflecting concerns about its long-term business model [00:43:17].
- A strategy for Google could be to drastically increase traffic acquisition costs (TAC) paid to publishers, making exclusivity deals to prevent their content from being used by competing AI agents [00:45:16]. Google currently pays Apple an estimated $15 billion annually to be the default search engine on iPhones [00:52:17].
AI and Intellectual Property Challenges
The core of the AI and intellectual property issues debate revolves around data rights and fair use:
- Attribution and Compensation: When AI synthesizes information from numerous sources, it’s difficult to attribute the output to specific content publishers or pay them royalties [01:21:40].
- Fair Use Doctrine: The concept of “fair use” in copyright law allows for the creation of “transformative” works that add “new expression or meaning” to the original [01:02:17]. However, concerns arise if AI merely rewrites content without transforming it, potentially harming the original copyright owner’s ability to profit [01:04:05].
- Legal Battles:
- Getty Images vs. Stable Diffusion: Getty Images is suing Stable Diffusion for training its AI on Getty’s watermarked images, leading to allegations of copyright infringement [01:03:59].
- GitHub Copilot Lawsuits: GitHub Copilot, which generates code, is facing lawsuits from developers for allegedly being built on “stolen content” from open-source projects [01:30:20].
- Content Publisher’s Dilemma: Publishers worry that AI summarization might interfere with their ability to monetize their content [01:00:01]. Some argue that content providers should “unionize” to demand payment for their data, similar to the music industry’s collective bargaining [01:22:59].
- Open Internet vs. Closed Apps: While the internet traditionally allows open access to content, increasingly, content resides within closed apps (e.g., Facebook, Instagram) [00:55:04]. The question arises whether exclusive deals with publishers (like Quora) for their content could create competitive moats for AI models [00:57:38].
Future Outlook
Despite the significant legal and cost issues, the advancement of AI is seen as inevitable. Technology waves this powerful are typically not “stymied by either chip costs or legal rights issues” [01:02:57]. Instead, these issues are expected to be “sorted out” through litigation and licensing fees, similar to how YouTube addressed piracy by enabling content creators to watermark and monetize their content [01:02:54].
Regulation and oversight of AI in this space is a growing concern. The debate also highlights that while AI can automate content creation, the human element of adding new intelligence and value remains crucial [01:05:02]. The expectation is for more startups and niche products to emerge, leading to “more innovation” and ultimately “greater economic productivity” for the end customer [01:09:22].