Podcasts

This podcast marks the final interview with Manus' co-founder and chief scientist, Ji Yichao, before its full acquisition by Meta. Ji Yichao recounted his experiences, from successfully monetizing personal projects via the APP Store in high school to becoming one of China's first-generation software entrepreneurs to venture overseas. He shared his journey from mobile browsers to Natural Language Processing (NLP), discussing the research and challenges in knowledge graphs, semantic search, and open information extraction technologies. The podcast delved into the profound impact of GPT-3 on the AI field and the critical balance between technology-driven and product-driven approaches in entrepreneurship. Notably, he elaborated on how the Manus team pivoted from an AI-native browser project to developing Matters, a cloud-based general AI Agent product, and secured cash flow and user insights through the Monica plugin. The interview also explored the distinctions between the AI industry and traditional mobile internet entrepreneurship, the competitive yet collaborative relationship between Large Language Models (LLMs) and application companies, team decision-making models, the future prospects of general Agents, the data flywheel effect, AI product internationalization strategies, and the paramount importance of mental and physical well-being in entrepreneurship. Ji Yichao emphasized that in the AI era, "what not to do" is often more crucial than "what to do," reflecting on the need for restraint and focusing on core product competitiveness. The ultimate goal is the widespread application of AI Agents in the real world, with high anticipation for the synchronized iteration of the Make Matters product and its underlying models in 2025.

This podcast features content creator and product manager Liu Fei. A self-described 'highly introverted individual' and someone with 'limited energy reserves,' he shares how, after leaving a tech giant, he developed a unique 'battery usage guide' – an energy and attention management strategy – to accomplish more valuable tasks in less working time. The core of his approach is to minimize unnecessary energy expenditure and cultivate sustainable motivation. He introduces a 'one-shot' working mode to reduce switching costs and cognitive residue, thereby avoiding the inefficiency often associated with Parkinson's Law. He advocates for 'completing today's tasks today' to mitigate emotional debt and emphasizes proactive decision-making and streamlined collaboration processes to combat decision fatigue. For 'recharging,' he posits that high energy is not innate but stems from continuous positive feedback and clear, specific motivations. He encourages experimentation in the 'gray areas' – tasks that are 'optional but potentially beneficial' – to unlock more possibilities. The podcast aims to inspire tech professionals who also feel energy-limited, helping them enhance both efficiency and job satisfaction.

This episode features Saul Perlmutter, the 2011 Nobel Laureate in Physics, delving into the application of scientific thinking in navigating an uncertain world. Perlmutter emphasizes that "skepticism" – viewed as critical inquiry – is a scientist's superpower, driving them to constantly seek out errors and foster progress. He elaborates on the concept of "confident humility," which involves balancing the acknowledgment of one's shortcomings with the belief in one's ability to solve problems. The podcast also introduces scientific methods such as "blind analysis" and "scenario planning" to counter cognitive biases and the "herd mentality," thereby enhancing decision robustness. Regarding team collaboration, Perlmutter highlights the importance of diverse skill sets and open communication, and encourages friendly competition as a catalyst for innovation. In the AI era, he believes critical thinking is paramount, as AI, being a double-edged sword, can both assist and mislead, thus requiring judicious use. Finally, from a cosmologist's perspective, he shares insights on the discovery of the accelerated expansion of the universe and the mystery of "dark energy," encouraging the younger generation to actively participate and collaboratively create an ideal world with constructive thinking, rather than succumbing to doomsday narratives.

In this episode, Howie (Silicon Valley Teacher Xu) and Dai Yusen from ZhenFund review the unexpected changes in the AI field in 2025 and look ahead to the opportunities and challenges of 2026. The discussion centers on the rapid catch-up of Chinese model companies, particularly the rise of Chinese power in the open-source sector. The guests explore the paradigm shift from 'scaling' to 'research' and the role of new AI Labs in driving innovation. Regarding Agent technology, they emphasize that its development is the 'beginning of a decade' rather than an overnight success, facing challenges in completion, reliability, and accountability. On commercialization, they compare mobile internet with AI entrepreneurship, noting that AI monetization models are still being explored, predicting potential disruption to traditional advertising, and analyzing subscription, advertising, and usage-based billing. The podcast also discusses 2026 as the 'Year of Delivering Results' for the AI industry, meaning pressure to meet expectations, and analyzes the dangers and opportunities for the Magnificent Seven in the AI wave, ultimately affirming the investment value of AI as a major technological shift while encouraging early-stage innovation.

In this podcast episode, Ren Xin and Xu Wenhao, co-hosts of AI Alchemy, delve into the major advancements, unexpected developments, and evolving perspectives within the AI domain in 2025. They emphasize that successful teams in 2025 were those adept at quickly leveraging scenario-specific benefits from model upgrades, rather than solely pursuing technological leadership. The podcast highlights AI's efficiency gains in areas like video production and advertising, predicting that 2026 will be a 'year of significant job displacement' as AI Agents truly achieve widespread adoption, shifting from delivering 'intermediate results' to achieving 'business objectives.' They discuss the impressive experiences offered by AI tools such as DeepSeek and O1 Pro, and AI's notable progress in code generation and visual creation. Concurrently, the podcast reflects on the current challenges of low penetration rates for AI applications in vertical sectors and the difficulty in changing user habits. Regarding entrepreneurship and organizational structure, they propose 'low-ego organizational culture' and the 'lean startup approach (unofficial theory),' advocating for a small team collaboration model characterized by high talent density and low psychological burden. Finally, the program offers practical advice for anxious individuals, practitioners, and entrepreneurs, encouraging them to embrace change, focus on solving concrete problems in traditional industries with AI, and pointing out structural arbitrage opportunities, such as the China-US market price difference.

This podcast focuses on the Manus platform, a general AI Agent platform defined as an 'Action Engine'. Guest Ivan Leo details how Manus has evolved from the traditional AI model of 'providing answers' to an automated model of 'executing tasks'. Through multiple live demos—including a French learning assistant, browser automation, and event data scraping for website building—the podcast demonstrates Manus's powerful versatility and flexibility. The episode dives deep into the design philosophy and core capabilities of the Manus API, covering sandbox environments, Docker image support, Stripe integration, Webhooks, and file handling mechanisms. The most compelling demo features building a Slack bot with Manus to automate a reimbursement process: from invoice recognition and querying Notion policies to updating reimbursement forms, the entire workflow requires no human intervention. The podcast also discusses Manus's privacy and security strategies, as well as its roadmap for long-term memory and browser API authorization, providing practical guidance and forward-thinking insights for developers and listeners interested in AI automation.

This podcast invited Wang Bingru, a U.S.-based reporter; Banfo Xiansen, a financial blogger; and Sanbiao Longmenzhen, a prominent self-media creator, to discuss how text and image creators perceive the significance and enduring nature of text amidst the current competition for traffic across various media forms. The guests shared their creative experiences on platforms like Toutiao, including their reflections on viral content (baokuan), the transition from data anxiety to a more balanced mindset, and how to balance personal creation with team operations. They specifically delved into the impact of AI technology on content creation, viewing AI as a tool to enhance efficiency but one incapable of replacing genuine human expression and deep thought. The three guests unanimously agreed that text, as the foundation of all content and a bridge connecting people's hearts, possesses irreplaceable long-term value. They expressed optimism about the future development of text and image creation, emphasizing that high-quality content and authentic personal voices are increasingly precious in the AI era.

This podcast delves into profound reflections and debates within the AI field. Meta Chief Scientist Yann LeCun (LeCun) posits that while current LLMs excel in language processing, they fall short of achieving human-level general intelligence due to a lack of deep understanding of the real world and efficient sample learning capabilities. He suggests that the JEPA architecture or World Model he champions could offer new directions. LeCun bluntly states that when he says “machine learning sucks,” it is not to negate the technology itself, but rather to highlight the vast disparity in learning efficiency compared to humans. DeepMind scientist Adam Brown, however, holds a more optimistic view, arguing that LLMs, by simply predicting the next word, have exhibited the emergence of complex intelligence and superhuman capabilities in fields like Math Olympiads. He emphasizes the significant potential for simple rules to generate complex behaviors at a massive scale and notes that AI capability growth shows no signs of saturation. The two experts also thoroughly discuss AI consciousness, autonomy, safety and controllability, and the societal value of open-source AI. LeCun frames AI safety as an engineering problem solvable through well-defined goals and guardrails; Brown, conversely, expresses concerns about the potential risk of “goal misalignment” in AI. Both acknowledge the critical role of open-source initiatives in preventing information monopolization and envision a future where AI enhances human intelligence and promotes societal progress.

This podcast brings together Anthropic's MCP project lead, an OpenAI representative, Block's chief engineer, and the Linux Foundation CEO to review the significant progress of the MCP Protocol in the year since its release. The content covers MCP's journey from an internal Anthropic tool to an industry standard, detailing its technological evolution and challenges in authentication, HTTP streaming, and tasks (Tasks) and UI interaction (MCP Apps). The guests also delve into the background, goals, and governance principles of the "Agent AI Foundation," emphasizing the importance of ensuring the protocol's openness and fostering cross-company collaboration through a neutral foundation during a period of rapid AI development. The podcast explores the relationship between Code Patterns, Skills, and MCP, as well as MCP's internal enterprise applications and future consumer-level vision, presenting listeners with a comprehensive picture of the Agent AI collaborative ecosystem.

This episode of the "Random Book Flipping" podcast delves into how the AMA (Ask Me Anything) model has surged on the Xiaohongshu platform, emerging as a localized practice of the "Build in Public" concept within the Chinese internet. Guests Professor Xu Huazhe and veteran internet observer Lan Xi analyzed the deep reasons behind AMA's popularity: first, it addresses the demand driven by information scarcity, as people in the era of Large Language Models increasingly crave authentic human conversations; second, the platform's unique vertical content distribution mechanism cultivates Xiaohongshu as a space for high-quality, non-performative dialogue. The podcast posits that AMA breaks down traditional information barriers, promoting the equal dissemination of knowledge, especially in high-barrier fields like AI. It serves as an effective channel for scholars and developers to engage in public exchange and receive feedback. Additionally, AMA offers entrepreneurs a valuable opportunity to 'seek and integrate advice,' accelerating product iteration and the establishment of trust capital. Finally, the guests explored AMA's significance for the evolution of the Xiaohongshu community, believing it introduces high-concentration human capital, expands the platform's commercial boundaries and influence, and positions it as a core incubator for AI technology and new trends.