ai
At this moment of inflection in technology, co-hosts Elad Gil and Sarah Guo talk to the world's leading AI engineers, researchers and founders about the biggest questions: How far away is AGI? What markets are at risk for disruption? How will commerce, culture, and society change? What’s happening in state-of-the-art in research? “No Priors” is your guide to the AI revolution. Email feedback to show@no-priors.com. Sarah Guo is a startup investor and the founder of Conviction, an investment firm purpose-built to serve intelligent software, or "Software 3.0" companies. She spent nearly a decade incubating and investing at venture firm Greylock Partners. Elad Gil is a serial entrepreneur and a startup investor. He was co-founder of Color Health, Mixer Labs (which was acquired by Twitter). He has invested in over 40 companies now worth $1B or more each, and is also author of the High Growth Handbook.
Jake Stauch, co-founder and CEO of Serval, discusses how his enterprise automation company has scaled rapidly from 20 employees in January to 200 today. He explains that Serval is built to directly challenge legacy enterprise software providers like ServiceNow by using code-generation agents combined with strict enterprise guardrails and approval workflows.
Jake Stauch reveals that Fox Corporation is one of Serval's largest customers, having adopted the enterprise automation platform company-wide. According to Stauch, the employee who led the migration on the ground level was nominated as the MVP of the entire company for his work.
Dane Knecht states that vulnerabilities reported in CVEs are increasing exponentially and expects that the majority of engineering teams' time will be spent on security over the next 18 months. He notes that keynote data indicates over 50% of engineering time is shifting toward security during this transition period.
Dane Knecht discusses how the attacker landscape has evolved, stating that recent major vulnerabilities are not just from nation-state actors but also from high school students dubbed 'agent kitties' who point AI agents at targets and leverage bug bounties.
Dane Knecht announces that Cloudflare has released a content monetization gateway with agent wallets and a stablecoin to facilitate microtransactions for content creators, alongside a project called Signals with OpenAI.
Dane Knecht shares that Cloudflare hired 1,000 interns over the course of the year for their 5,000-person company because younger cohorts are fully AI-native and help accelerate the internal adoption of AI tools.
Dane Knecht explains that Cloudflare plans to expand its red teams from one team to four or five groups by the end of next year to act more adversarially and actively discover software vulnerabilities.
Walter Goodwin, CEO of Fractile, explains the company's strategy of controlling the entire AI chip design process, from architecture to physical design. This full-stack approach allows Fractile to be more agile and responsive to the rapidly evolving AI workload landscape, contrasting with traditional, more siloed chip development methods.
Fractile is focusing on achieving significantly higher bandwidth to memory in their AI chips, moving away from traditional SRAM-based designs. This shift is driven by the need to support increasingly long context lengths in AI models and improve efficiency, especially for tasks like running large Mixture of Experts (MOE) models.
Walter Goodwin observes that while many AI ASICs currently appear similar, relying on common components like HBM and tensor cores, there's a growing need for differentiated accelerators. Companies like Fractile aim to offer new capabilities, particularly in memory bandwidth, to address evolving workload demands.
Walter Goodwin discusses the tension between the need for rapid AI chip development cycles and the necessity of making long-term architectural bets. While compressing the design cycle offers more 'shots on goal,' chips still need a multi-year useful lifespan, requiring careful consideration of future workload evolution.
Walter Goodwin believes the timeline for fully automating chip design from intent to GDS2 file is shorter than industry predictions. He highlights that while AI can automate many design tasks, computational bottlenecks in areas like layout and simulation will persist, requiring continued innovation.
Fractile aims to create a significant competitive advantage by developing AI chips that are structurally 'ahead' of the market, potentially carving out a three-to-six month lead. This strategy, according to CEO Walter Goodwin, is crucial for winning deployments in the fast-paced AI hardware sector.
Sequence Holdings, led by co-founder and CEO Michael Lee, has announced the acquisition of Baldwin, an insurance broker, in a take-private transaction valued at $7.7 billion. This deal marks the largest AI-focused take-private transaction to date.
Michael Lee, CEO of Sequence Holdings, describes the company's mission as partnering with management teams of established businesses to integrate AI and transform them into market leaders. The firm focuses on acquiring and refunding businesses with a long-term vision for technological integration.
Michael Lee explains his perspective that AI will have an uneven impact on the economy, affecting some industries more than others. He believes that by acquiring established companies with inherent advantages like brand or scale, and then integrating AI, it's possible to forge new market leaders.
Michael Lee likens the current AI advancements to the next industrial revolution, emphasizing the need to reorganize organizations to leverage technology's full potential. He contrasts this with simply providing employees with AI tools, advocating instead for a fundamental rethinking of organizational structures to adapt to AI's capabilities.
Stefano Ermon, co-founder and CEO of Inception, reflected on his research journey starting in 2014 at Stanford, focusing on generative models. He noted the early limitations of models, like those used for MNIST digit generation, and the shift in focus towards their utility in learning features from unlabeled data.
Stefano Ermon explained that diffusion models originated from score-based generative models, a concept he and his PhD student Young Song developed in 2019. The core idea was to train a neural network to denoise images, which could then be used to generate new images by starting from noise and gradually refining it.
Stefano Ermon highlighted a 2024 breakthrough where diffusion models matched the quality of autoregressive models like GPT-2 for text generation. Crucially, these diffusion models were significantly faster, achieving text generation speeds up to 10 times quicker due to their parallel processing capabilities.
Stefano Ermon founded Inception to scale up diffusion model technology for commercial language models. He believes diffusion models offer a competitive advantage due to their efficiency at inference time, contrasting them with the sequential nature of autoregressive models.
Stefano Ermon argued that diffusion models are fundamentally better than autoregressive models for inference time efficiency. He drew an analogy to the shift from RNNs to transformers for training parallelism, stating that diffusion models enable similar parallelism for inference, making them more suitable for GPUs.
Brian Armstrong, CEO of Coinbase, outlines the company's vision for the 'Everything Exchange,' which aims to consolidate trading for all asset classes, including stocks, commodities, and crypto. He highlights stablecoin payments as a rapidly growing area, enabling fast and cheap global transactions.
Coinbase is developing 'agentic finance' to provide financial services for AI agents, allowing them to have their own accounts and conduct transactions. This initiative aims to support the growing agentic economy, which is predicted to eventually surpass the human economy in size.
Brian Armstrong highlighted that a significant majority (76%) of agentic commerce transactions are valued at less than 30 cents. This contrasts sharply with traditional payment methods like debit or credit cards, which have a minimum fee of approximately 30 cents, making them inefficient for such small transactions and underscoring the need for new payment solutions.
Brian Armstrong believes that AI and crypto rails are crucial for the future of the agentic economy. He suggests that even with potential price deflation due to AI and robots, scarcity of certain resources like land and energy will persist, necessitating a medium of exchange. He views crypto as a natural complement to AI for these future transactions.
Coinbase is integrating AI into its operations, starting with basic tools like coding assistance and expanding to machine learning for risk and fraud prevention, and customer support. The company is also exploring more advanced applications, such as creating an internal 'company brain' for recursive self-improvement.
Arm CEO Rene Haas highlighted the significant role of AI in accelerating chip design and verification processes. He stated that AI is particularly effective in verification, validation, and debugging, which are the most time-consuming parts of chip development. Haas also noted that a large percentage of Arm's engineers use AI tools daily, likening its absence to a severe limitation.
Arm is evolving beyond IP licensing to directly producing physical chips, beginning with their Arm AGI CPU launched last March. This move into chip manufacturing requires new capabilities in supply chain operations, working with foundries like TSMC and Samsung, and memory allocation with suppliers such as Micron and SK Hynix.
Arm's entry into producing physical CPUs was partly driven by a request from Meta, which sought a general-purpose AGI CPU that was not available on the market. Meta approached Arm to collaborate on this, marking a significant partnership for Arm's new venture.
Arm received significant support from major tech companies like Nvidia, Amazon, Microsoft, and Google for its expansion into producing physical chips. Haas mentioned that these companies, which build Arm-based server chips, were generally on board with the new direction.
Arm is actively working with model makers to fine-tune AI models, particularly for areas like RTL generation and physical design where current tools are less mature. Haas believes Arm's extensive IP portfolio, including documentation and test benches, provides a competitive advantage for training AI models.
Ofir Erlich and Gonen Stein, co-founders of Eon, emphasized that in the current AI-driven landscape, data is the most valuable asset for organizations. They contrasted this with compute and models, which they described as ephemeral with low switching costs. This perspective was highlighted by Google's recent $10 million acquisition of Spirit Airlines' data.
Ofir Erlich and Gonen Stein of Eon observe a significant trend of companies seeking to acquire data, including from entities in bankruptcy. They note that data, often previously stored as tapes and collecting dust, is now recognized as a critical asset for training AI models and gaining a competitive advantage. This shift is driven by the realization that data is gold and can be leveraged to build business value in the AI era.
Eon provides a cloud-based data foundation designed for the AI era, enabling companies to map, classify, and ingest their data from various sources. According to co-founder Ofir Erlich, this foundation makes data cost-effective to maintain for protection and recovery, while also allowing easy access and querying for AI model application.
The nature of security threats is evolving, with non-human actors now posing risks that are amplified compared to traditional human threats. Fortunately, the detection and protection methodologies remain similar, but the velocity at which these threats can occur is extreme, according to an unidentified speaker.
Gonen Stein from Eon suggests that AI is leveling the playing field for businesses, allowing both large and small companies to compete more equally. He posits that while people remain a crucial asset, the unique data a company possesses is now the primary advantage it holds.
Erik Allebest, CEO of Chess.com, shared impressive growth figures for the platform, stating they have over 250 million registered members and are projected to exceed $200 million in revenue this year. The platform currently sees about 10 million daily active users and 40-50 million monthly active users.
Erik Allebest explained that Chess.com was built without external funding, instead growing organically through revenue generated from memberships. He highlighted a strong focus on user experience as key to their success, differentiating them from competitors.
Erik Allebest discussed the significant growth periods for Chess.com, noting a major surge after the Netflix show 'The Queen's Gambit' and another wave in 2023 due to increased school play driven by short-form content and a cheating scandal. He mentioned that the platform's growth has remained high and compounding.
Chess.com CEO Erik Allebest recounted the company's financial journey, noting that while they largely bootstrapped, they did bring in investors starting around 2020-2021. He explained that these were primarily secondary investors, and the company still considers itself bootstrapped.
Erik Allebest stated that Chess.com's current mission is ambitious: to get a billion people to play chess. He believes chess offers significant benefits, such as bringing people together, building confidence, and teaching valuable life lessons about learning from losses.
Melisa Tokmak, founder and CEO of Netik, discussed how her company is using AI to manage millions of real-world businesses, focusing on essential services like HVAC and pet care. Netik aims to streamline customer interactions and optimize operations for large enterprises.
Melissa Tokmak highlighted the unreliability of human customer service teams in large, essential service businesses, citing issues like high staff turnover and no-shows. She explained that Netik's AI aims to provide a consistent and optimized customer experience, especially during peak demand.
Netic's AI agents are designed to manage customer interactions across various channels, including voice and text, for businesses in sectors like HVAC and home services. These agents handle complex queries, assess customer needs, and optimize service deployment.
Melissa Tokmak explained that Netik's AI is crucial for businesses in essential services due to the complexity of their operational needs. The AI helps in understanding customer requirements, determining serviceability, and optimizing deployment to enhance customer satisfaction and company revenue.
According to Melisa Tokmak, companies previously relied on large human teams for customer support, which hindered growth due to the need for constant investment in people. Netik's AI aims to replace these teams, allowing businesses to focus on core labor and improve efficiency.
Glenn Fogel, CEO of Booking Holdings, shared details of his unconventional start in the tech industry, beginning as a mainframe operator at Morgan Stanley in the early 1980s. He explained how this experience, far removed from his finance background, ultimately led him to pursue law and then a career on Wall Street.
Glenn Fogel, CEO of Booking Holdings, emphasized that competitive advantages and moats are temporary. He believes the only sustainable long-term strategy is continuous development of new services and ways of doing business to better serve customer needs.