No Priors Ep. 71: The Best of 2024 (so far) with Sarah Guo and Elad Gil

Too Long; Didn't Watch — Summary

This episode of No Priors reviews the first half of 2024, featuring highlights from interviews with prominent AI leaders discussing the transformative impact of AI across fintech, creative design, robotics, generative video, software engineering, and the crucial role of data quality and evaluation in building trust and advancing AI systems.

Main Takeaways

• AI is revolutionizing fintech by enhancing identity verification, regulatory compliance, and enabling seamless financial integrations for businesses.
• AI will augment human creativity in design, shifting collaboration from human-to-human to human-to-AI, leading to more and better-designed software.
• Developing complex hardware and software, like humanoid robots, requires an iterative design approach focused on rapid prototyping and continuous updates.
• Generative video models like Sora empower artists to tell stories more easily and are expected to lead to entirely new forms of media and advanced world simulations for applications like robotics.
• The future role of human software engineers will evolve towards technical architecture and product management, focusing on problem-solving and foundational computer science knowledge, rather than just coding.
• Building trust in AI systems is paramount, requiring robust data quality, continuous evaluation, and public transparency through rigorous benchmarks and leaderboards.

Detailed Summary

Introduction (0:00 - 0:46)

The episode serves as a mid-year review for 2024, highlighting key moments from previous interviews that cover the state-of-the-art in AI research, hyperscalers, and upstarts. The hosts mention that links to the full episodes are available for listeners to revisit the complete conversations.

Emily Glassberg Sands on the Future of AI and Fintech (0:46 - 4:22)

• Emily Glassberg Sands, Head of Information at Stripe, discusses the intersection of fintech and AI.
• She emphasizes the importance of AI in understanding business identity, what they sell, and navigating complex regulatory environments for credit, lending, and supportability decisions.
• AI can enable more seamless financial integrations, potentially reducing the need for extensive payments engineers or complicated developer work.
• Large Language Models (LLMs) are already proving effective at writing code for financial integrations.
• Stripe's business model aligns with helping users grow, suggesting that using financial data to improve businesses and grow the GDP is a powerful application of AI.

Dylan Field on AI and Human Creative Potential (4:23 - 9:02)

• Dylan Field, founder of Figma, discusses how AI is changing the design process and bridging the gap between design and development.
• He notes a shift from human-to-human to human-to-AI collaboration in creative fields.
• AI can augment creativity by assisting with iterative design loops and brainstorming, helping designers get past the initial blank canvas.
• Field believes AI will augment, rather than replace, designers, leading to increased efficiency and access to design tools.
• He predicts an increase in the amount and quality of software created, with engineers spending more time on design-related tasks as coding abstraction levels rise.

Brett Adcock on Running Figure AI’s Hardware and Software Processes (9:03 - 12:42)

• Brett Adcock, CEO of Figure AI, discusses the challenges and strategies for developing humanoid robots that combine hardware, software, and AI.
• Figure AI employs an iterative design approach, prioritizing rapid testing and prototyping over extensive research.
• The company aims for continuous updates to both hardware and software, acknowledging that perfection is an ongoing pursuit.
• Their development process starts with understanding customer needs, setting detailed requirements (e.g., lifting capacity, battery life, safety), and then moving through conceptual, preliminary, and critical design reviews.
• Unlike software development, hardware development involves much longer timelines and requires a more methodical approach to design from scratch, including trade studies to avoid designing non-functional components.

OpenAI’s Sora Team on Artists’ Creative Experiences with their Model (12:43 - 17:42)

• The OpenAI Sora team discusses the generative video model and its impact on creative content creation.
• They highlight how artists are using Sora in unexpected and brilliant ways, such as Shy Kids' short story "Airhead."
• Sora enables artists to tell stories more easily, making complex video creation accessible even to those without traditional skills, as exemplified by the "Bling Zoo" sample.
• The team anticipates that Sora will not only lead to more traditional short films but also inspire completely new modes of interacting with content.
• They also discuss Sora's potential as a world simulation model, acting as a physics engine for simulations and a basis for robotics by understanding physical world dynamics from raw video data.

Scott Wu Gives Advice for Human Engineers Co-Working with AI (17:43 - 21:05)

• Scott Wu, co-founder of Cognition (developer of Devin, an AI engineer), discusses the evolving role of human software engineers in an AI-augmented future.
• He acknowledges the meme that "the hottest new programming language is English" but stresses the enduring value of software engineering fundamentals, such as understanding computer internals, logic gates, algorithms, and logical reasoning.
• Wu predicts that in 5-10 years, a software engineer's role will resemble a mix of a technical architect and a product manager, focusing on problem breakdown and solution design.
• He emphasizes that while AI's long-term impact on super-intelligence is hard to predict, its concrete effects on work and daily life are already being seen, with much more innovation to come.

Alexandr Wang on How Quality Data Builds Confidence in AI Systems (21:06 - 25:20)

• Alexandr Wang, founder of Scale AI, discusses the critical role of data quality and evaluation in building trust in AI systems as they approach and surpass human abilities.
• Scale AI focuses on enhancing the entire AI lifecycle, from ensuring data abundance and quality to measuring AI systems to build confidence and enable further development.
• Evaluation and measurement are crucial for governments, enterprises, and labs to ensure AI systems are safe, secure, fit for adoption, and developed responsibly.
• Wang explains that evaluating AI is challenging because it's like building IQ tests for models, and current academic benchmarks often suffer from models being overfit or having training data leakage.
• Scale AI introduced DSM 1K, a held-out evaluation for math capabilities, which revealed discrepancies between reported and actual model performance.
• He advocates for leveraging human experts for rigorous evaluations and calls for public visibility, transparency, and leaderboards to demonstrate AI performance scientifically.

Conclusion (25:20 - 25:52)

• The hosts express gratitude to listeners and the guests who are reshaping the world with AI.
• They encourage listeners to find full episode links in the description and to subscribe for new weekly interviews.
• Listeners are invited to follow No Priors on Twitter, YouTube, Apple Podcasts, Spotify, or their preferred podcast platform, and to sign up for emails or find transcripts at no-priors.com.

Notable Quotes

"having a really robust understanding of identity who businesses are are what they're selling um has always been important and you know I think often in Industry we think it's important for marketing or sales or sort of go to market motions but it's also super important in fintech" — Emily Glassberg Sands

"you're moving from a human to human collaboration company to a human to AI collaboration company over time" — Elad Gil

"I think design is actually in a really good place probably before you see potential replacement of any part part of the design role you instead see augmentation and you see access uh you see efficiency so that designers can get more done" — Dylan Field

"we want to be continuously updating the hardware and software forever it'll I don't think it'll ever be good enough for us" — Brett Adcock

"it was really cool to see a way that Sora can unlock and make this story easier for them to tell and I think there it's even less about like a particular clip or video Sor made and more about this story that the these artists want to tell and are able to share and that Sora can help enable that" — Aditya Ramesh

"I totally think that carrying out simulations in the video model is is something that we're going to be able to do um in the future at some point" — Aditya Ramesh

"I think the role of a software engineer um five or 10 years from now looks something like a mix between a technical architect and a product manager today" — Scott Wu

"evaluation and measurement of the AI systems is a critical component of the life cycle but also a critical component I think of of society being able to build trust in these systems" — Alexandr Wang

"grading one of these AI systems is is not something that's very easy to do automatically and it's it's sort of like um you know you have to kind of build IQ tests for these models" — Alexandr Wang