No Priors Ep. 59 | With Sarah Guo & Elad Gil

Detailed Summary

Introduction (0:00 - 0:32)

The hosts, Sarah Guo and Elad Gil, introduce a host-only discussion to take a step back and analyze the significant changes occurring in the AI landscape over the past few weeks.

The Evolving Model Landscape (0:32 - 3:21)

• The landscape of AI models is rapidly changing, with expectations that several GPT-4 level models, including open-source options, will be available by year-end.
• Companies like Mistral and DataBricks (with DBRX) are demonstrating what can be achieved with relatively smaller compute resources.
• DataBricks introduced "Mosaic's Law," suggesting that the capital investment for a given model capability will decrease by 75% annually due to hardware and algorithmic improvements.
• While an oligopoly might still exist at the cutting edge due to capital scale, the broader availability of capable models is beneficial for application development and enterprise adoption.
• Advanced models can bootstrap the creation of future models through data labeling and reinforcement learning, accelerating progress.
• The cloud providers are poised to capture significant value by hosting these diverse models.

Microsoft/Inflection Deal (3:21 - 5:28)

• The Microsoft/Inflection deal is seen as a strong affirmation of Microsoft's commitment to AI and a strategic move to acquire top AI talent and leadership.
• This move addresses Microsoft's need for AI-aware product and research leadership across its properties, including Co-pilot, productivity suites, and search.
• While Microsoft continues its partnership with OpenAI, the Inflection deal is also viewed as a hedge, reducing reliance on an external player for a critical technological driver.
• The deal highlights the immense capital and compute resources required for state-of-the-art AI research, which independent players often struggle to match.

Voice Cloning and Multimodal AI (5:28 - 7:02)

• OpenAI's announcement of voice cloning capabilities raises questions about its market impact, especially given existing players like ElevenLabs.
• Hesitancy in releasing advanced voice AI has been attributed to regulatory and societal concerns, particularly regarding deepfakes.
• Solutions like attestation (where the voice owner verifies the upload) could mitigate some of these concerns.
• Many players likely possess advanced voice, video, image, and song generation technologies but have held back due to these concerns.

AI Investment Climate and Whitespace (7:02 - 12:50)

• Venture capitalists provide initial funding (tens to hundreds of millions), but the bulk of AI funding comes from hyperscalers and large tech companies (billions).
• Hyperscalers have a strong incentive to fund AI companies as it drives significant cloud revenue; for example, Microsoft's Azure revenue grew 5% due to AI-related products.
• While new language models might see less VC funding, significant opportunities exist in other foundation model areas like music, video, image generation, physics, biology, material science, and robotics.
• Elad Gil notes a "whitespace" in AI, with many obvious areas still unexplored by numerous smart teams.
• Areas like operating on time series data with general knowledge (e.g., anomaly detection in infrastructure, security, healthcare) present huge commercial applicability.
• Biotech (e.g., protein folding with long context windows) and robotics are seeing breakthroughs, with smart teams applying foundation models to traditional benchmarks.
• Robotics faces challenges in data collection (embodied action data vs. internet data) and defining the value chain (software for diverse hardware vs. verticalized companies).

Mimetic Behavior and Commercial Applications (12:50 - 16:36)

• The concentration of smart, talented individuals on a few specific AI areas, despite vast whitespace, is attributed to a mix of perceived societal significance and "mimetic actions" or "memes" of what to build.
• Historical examples like early photo apps or search engines illustrate how many similar products emerge before a dominant one (e.g., Instagram, Google) with the right product substantiation.
• The current AI market is heavily driven by technologists, leading to a focus on intellectually interesting problems, which may not always align with immediate commercial value.
• There's a recognized gap in cross-pollination between AI researchers and domain experts in fields like finance and accounting, where AI could offer significant commercial capabilities.
• Recent technological breakthroughs have made previously intractable problems in various fields more feasible, driving excitement and development.

AI Video Landscape (16:36 - 19:54)

• The video AI landscape is more complex than just a new modality; controlling video for commercial applications differs from general video generation.
• Demand for high-quality, commercially viable, and shareable video generation is unbounded, but significant challenges remain in interfaces, controllability, length, and quality.
• Companies like HeyGen, an investor-backed company, are focusing on specific video applications like avatars for prosumer and commercial use, demonstrating rapid growth.
• The key to successful video applications lies in simplifying creative dimensions and lowering quality requirements in certain areas to make creation easier for users.

Agentic User Experiences (19:54 - 22:21)

• Devin by Cognition has highlighted the potential of agentic systems and new model user experiences.
• Devin's UI, with tabs for plan, shell, code, and chat, allows users to monitor and redirect the agent's actions, treating it like a "junior intern" that needs guidance.
• This transparent, interactive UI is becoming a new paradigm for agentic products, enabling users to interfere and guide the agent for faster and more accurate results.
• This human-in-the-loop approach is an intermediary step; eventually, as agents become smarter through base model and reasoning breakthroughs, the need for such extensive UI may diminish.

Prosumer as the First Wave of Application AI (22:21 - 25:42)

• Prosumer applications (e.g., HeyGen, ChatGPT, Perplexity) are leading the AI wave due to their ability to carve a path of direct user value and faster adoption compared to enterprise solutions.
• Enterprise adoption is slower due to complex decision-making, security processes, and roadmap integration.
• Prosumer products, often costing around $10/month, offer immediate productivity gains, leading to quicker user decisions.
• Canva's success, with $1 billion of its $1.7 billion ARR from prosumers, exemplifies the significant market potential of this segment.
• The current AI wave, starting with consumers and prosumers before moving to B2B, parallels the early internet and mobile waves, suggesting this dynamic is characteristic of fundamental technological shifts.

Notable Quotes

"It's very likely at this point that you end this year with a handful of GPT 4 level models and that some of those are open source." — Sarah Guo

"The most cutting-edge smartest models in some sense you're going to end up with an oligopoly at least in the next couple years just because of the scale of capital needed." — Elad Gil

"It's also hard from the outside not to see this as somewhat of a hedge, right? Not in a criticism of OpenAI, but you know, if you are a true believer that this is the most important technical driver for your company and then you're reliant on an outside player, that's not a position that a trillion-dollar company likes to have." — Sarah Guo

"Most of the funding of this market is actually being done by the hyperscalers and a few other big tech companies." — Elad Gil

"There's so much white space in AI right now, there's so many different things to do. I've actually started incubating a few things again just because there's, you know, people just aren't working on certain areas that seem kind of obvious." — Elad Gil

"The way to think about agents today I feel is almost like a junior intern. They're very eager, they're trying really hard to please, but they still have a lot to learn or you kind of need to give some direction." — Elad Gil

"It is very, very hard to get to millions of Enterprise users in a year simply because of the decision-making and like security processes and roadmap involved in getting a large customer to change something internally." — Sarah Guo

"My sense of the market is that multiple players have this technology but they've been holding it back and in some cases they may have had it for a year or two now." — Elad Gil

"We still have a very, very long way to go both in terms of interfaces and controllability, like length, quality, etc." — Sarah Guo