finance
Industry veterans, degenerate gamblers & besties Chamath Palihapitiya, Jason Calacanis, David Sacks & David Friedberg cover all things economic, tech, political, social & poker.
Jacob Coxin, an AI researcher formerly at OpenAI and most recently at Anthropic, resigned citing fears that AI could lead to human extinction by the end of the decade. He posted his concerns on X, which went viral, attracting significant attention and sparking debate about AI safety and regulation.
Evan Binger, who leads alignment science at Anthropic, confirmed the severity of AI existential risk fears, stating he personally believes there's "over 10%" chance of AI killing humans within the next decade. He also admitted that the company "do not yet have a plan to solve alignment."
The viral claims about AI existential risk have prompted reactions from politicians, with Bernie Sanders proposing legislation to ban superintelligence and pause AI development. Governor JB Pritzker echoed these concerns, calling for increased action to "reign in AI."
David Sachs expressed skepticism regarding the AI doomsday narrative, labeling it as "doomer, histrionics" and suggesting the recent viral posts are part of an orchestrated operation. He pointed to the lack of concrete evidence and the rapid amplification by specific groups as indicators of a coordinated effort.
David Sachs detailed a potential 'op' behind the viral AI doomer claims, citing the anonymous researcher's account having no prior activity, the rapid amplification by funded doomer groups like Encode AI and AI Policy Network, and a Wall Street Journal article appearing before the tweet storm as evidence of orchestration.
Eric Weinstein states he is in Washington D.C. to meet with figures like OSTP Director Kratzios and NIH's Jay Bhattacharya to review the first major proposal to reform American science since Vannevar Bush post-WWII. He believes this reform effort is critical and expresses optimism about the direction, noting Kratzios's growth in the role and a willingness to acknowledge the limits of current knowledge.
Eric Weinstein argues that America's strength in science lies in its capacity for 'cowboy science'—taking on courageous, high-risk bets that other nations cannot afford. He believes this freedom to question established dogma, unlike in China, is a significant advantage. He criticizes the current system for rewarding incremental research and penalizing those who challenge consensus, hindering the pursuit of truth.
Eric Weinstein asserts that the scientific community has become politicized, creating a 'scientific precariat' where professors fear challenging consensus due to funding and social pressures. He uses the example of COVID-19, questioning why skeptics were absent during the pandemic and arguing that the scientific method, which relies on falsification, was abandoned in favor of calcified narratives.
Eric Weinstein criticizes government-mandated policies during the COVID-19 pandemic, calling them unprecedented and contrary to the scientific method. He argues that the calcification of narratives and the punishment of dissenting voices, rather than rigorous hypothesis testing and falsification, characterized the pandemic response.
Eric Weinstein posits that scientists possess powerful tools to reveal the truth, even within covert operations. However, he believes that 'B minus policy people' in Washington, particularly within the covert community, prefer to silence independent scientists whose findings might contradict their narratives. This, he argues, extends to journalism, which also falls in line with a dominant narrative.
Eric Weinstein draws a parallel between the lack of reporting on President Roosevelt's health during WWII and the handling of President Biden's cognitive state. He questions why leading geriatric neurologists were not consulted to assess Biden's gait and speech patterns, suggesting a deliberate suppression of information when a 'lie' is deemed structurally important.
Google is experiencing significant shifts in its AI leadership, with Demis Hassabis moving to a new role and several top researchers, including Jeff Dean, leaving to start a new company called Discovery Loop. These departures reportedly stem from issues with Google's Gemini 3.5 Pro development and low morale within the AI division.
AI engineering legend Jeff Dean, a 27-year veteran at Google and employee number 30, has departed to co-found a new AI company named Discovery Loop. The startup is set to focus on "deep scientific breakthroughs in AI." This move follows news of other significant AI talent leaving Google.
David Friedberg suggests that Google's strategic shift towards AI infrastructure and data centers, rather than pure model development, is a key factor driving top AI talent like Jeff Dean to seek opportunities in startups. He highlights that infrastructure offers a more predictable return on investment compared to the high-beta nature of frontier model development.
According to Sachs, the AI market is clearly splitting into two tiers: a frontier intelligence duopoly (OpenAI and Anthropic) that can command premium pricing, and a commoditized market for six-to-12-month-old or lagging AI. Companies not at the cutting edge will struggle to charge for model layers, instead relying on compute and inference services.
Sachs highlights the dramatic growth of Anthropic, noting its Annual Recurring Revenue (ARR) has surged from $10 billion at the start of the year to an estimated $120 billion or higher by year-end. This rapid expansion is seen as evidence of the premium pricing power for frontier AI models.
Brad Gershner noted that a Morgan Stanley report indicates Microsoft is achieving over a 30% return on invested capital for its tokens-as-a-service business. This aligns with the broader trend discussed where infrastructure and services, rather than just model development, are proving highly profitable in the AI space.
Dino Mavroudis of Saronic detailed the use of their autonomous "Corsair" speed boat to rescue two pilots in the Strait of Hormuz. This mission marked the first time an autonomous ship was used for such a high-stakes rescue, demonstrating the Navy's trust in the technology with lives on the line.
Viv "Jason" Altacar and Dino Mavroudis discussed the stark contrast in shipbuilding capabilities between the US and China. Mavroudis stated that China can build 23 million gross tons of ships annually, compared to the US's 100,000 gross tons, a 230-to-1 ratio, with China holding 57% of global capacity.
Viv Altacar explained Saronic's mission to boost US ship production by focusing on autonomous vessels. He highlighted that removing human-centric components reduces complexity and cost, enabling faster, more efficient manufacturing compared to traditional ships.
Dino Mavroudis outlined the decline of the US naval fleet, noting it has 296 ships against a mandate of 355, while retiring more vessels than it builds annually. In contrast, China's navy has rapidly expanded over the last 30 years, now possessing nuclear submarines and aircraft carriers.
Dino Mavroudis explained how Saronic's 'Marauder' autonomous ships could revolutionize US naval capabilities by providing more firepower at a fraction of the cost of traditional destroyers. He stated they plan to build 20 Marauders per year, potentially fielding more Vertical Launch System (VLS) tubes than destroyers at a much lower price point.
Viv Altacar elaborated on the cost and efficiency benefits of autonomous ships, noting that removing human amenities like beds, commissaries, and bathrooms significantly reduces complexity and cost. This allows for greater optimization in naval architecture, speed, range, and payload.
Dr. Peter Funkhauser, co-founder and CEO of Anybotics, discussed the advantages of four-legged robot design for mobility and stability in hazardous environments. He explained that the quadrupedal format allows robots to navigate spaces similar to humans and maintain balance on uneven or slippery surfaces. Anybotics deploys these robots for industrial inspection, highlighting their ability to carry advanced sensors that can detect issues beyond human capabilities.
Dr. Peter Funkhauser of Anybotics explained that their robots are not designed to replace human labor but to perform tasks beyond human capabilities, particularly in inspection. These robots are equipped with advanced sensors like thermal cameras, acoustic microphones, and gas detectors to identify micro-leaks or overheating equipment that are imperceptible to humans. The value proposition lies in preventing costly downtime by identifying issues before they escalate.
Anybotics' robots are engineered to function in a wide range of environmental conditions, from extreme cold (-20 degrees Celsius) to high heat (+60 degrees Celsius), and can handle dust and humidity. Notably, the company has developed a robot specifically designed for explosive atmospheres, preventing any sparks to ensure safety in environments like oil and gas facilities where flammable gases like methane are present. These robots are crucial for inspections in such hazardous areas.
Anybotics CEO Peter Funkhauser stated that zero percent of their robot components are sourced from China. He explained this is due to historical sourcing practices favoring local and US-based chip manufacturers, which is important for some of their customers. The company has built much of its architecture in-house since its inception 10 years ago, emphasizing strategic sourcing of components globally to balance cost and performance.
When asked about robots being equipped to fix leaks while performing inspections, Dr. Peter Funkhauser clarified that while demonstrations exist, reliable automated repair in hazardous environments is still under development. He noted the current focus is on inspection and data collection, with the ultimate goal of having robots with multi-arm manipulation for repairs, but emphasized the significant work and AI advancements needed to achieve 99.99% reliability in complex, uncontrolled settings.
A former Intel executive believes the company's downfall was due to a shift in leadership from deeply technical individuals to business-focused executives. He argues that critical technical decisions, impacting billions, were no longer made by those with a strong engineering background.
A former Intel insider revealed that the company had not built a new factory in a decade and had forgone purchasing crucial EUV machines. He suggests these decisions, driven by short-term economics over long-term technological advancement, led to missed opportunities like providing chips for the iPhone.
During the five to six years preceding his return, Intel allocated $100 billion to shareholders through dividends and stock buybacks. The speaker implies this capital could have been better invested in technological upgrades, such as new factories and EUV machines, to prevent the company's decline.
The speaker posits that successful technology companies today are typically deeply technical and often founder-led. Even if not founder-led, leaders like Sunder Pichai are highlighted as being technically adept, emphasizing the importance of technical decision-making over purely spreadsheet-driven analysis for large-scale investments.
A company building a communication platform for AI has achieved $600 million in ARR, with revenue growing rapidly after releasing its human-sounding text-to-speech model in early 2023. The company has also maintained zero attrition among its founding research and engineering team of 10 people.
A tech company is embedding engineers into non-engineering teams, such as talent, legal, and revenue, to drive automation and ensure AI adoption and security. This approach is part of their strategy to build a communication platform for AI.
A rapidly growing tech company has eliminated the role of product managers, opting instead to leverage AI to summarize calls, create action items, and manage Kanban boards. The company believes this structure, combined with cross-functional teams and embedded engineers, optimizes product development.
The nature of software development has drastically changed with the rise of AI, enabling individuals without traditional coding backgrounds to build production code. This democratization of development empowers non-developers and increases the productivity of existing developers.
Early speech-to-text technology like 'Dragon Dictation' was hampered by a perceived awkwardness of talking to a computer, leading users to do so privately. Modern AI advancements, exemplified by products like Whisperflow, have made this interaction more natural and integrated, even using wearable devices.
Andrew Feldman, CEO of Cerebras, stated that data centers being built for AI in the coming years will consume more power than was used globally in the previous 50 years. He noted that these centers are physical structures with 'enormity' and are being constructed across the US, Nordics, Europe, Middle East, and even in countries like Kazakhstan and Tajikistan.
Andrew Feldman of Cerebras revealed that major AI companies like OpenAI, Anthropic, and Google are placing orders for chips years in advance, leading to a $25 billion backlog for Cerebras. He characterized this demand as 'insatiable' and noted that companies are trying to capture 'yesterday's demand' due to the inability to build data centers and hardware fast enough.
The capabilities of AI models have significantly advanced, moving from basic web result summarization to understanding user intent within 24 months. This evolution means users no longer need to be 'prompt whisperers' as AI can infer what is wanted, even suggesting better solutions than initially requested.
The speaker described interacting with an AI agent that exhibited reasoning by debating optimal strategies for finding global trends, considering sources like Hacker News, Reddit, and Instagram. This 'reasoning model' worked in the background to identify emerging trends, indicating a significant leap in AI's analytical capabilities.
The concept of 'unlimited tokens' in AI is seen as a gateway to 'unlimited reasoning,' significantly accelerating the thinking process for complex tasks. By leveraging faster hardware like Cerebras, what might take weeks or months of human thought could be achieved in hours, enabling unprecedented levels of discovery and insight.
Palantir and Nvidia have announced a partnership to develop a sovereign AI operating system for the US government. This collaboration will leverage Nvidia's AI technology with Palantir's platforms to build custom AI models, giving government agencies control over their hardware, data, and model weights.
Palantir CEO Alex Karp expressed dissatisfaction with frontier AI models, suggesting they lead to a loss of trust and a waste of enterprise resources. He argues that companies risk losing their intellectual property to these models, which may then compete with them.
Chamath defined 'intelligence sovereignty' as the concept of not allowing AI to dictate one's thinking or interpretation of the world, distinguishing it from mere privacy. He emphasizes the importance of owning data and models, advocating for open-source solutions and local hardware.
David Sacks highlighted that enterprise customers prioritize control over their compute, models, data, and proprietary knowledge ('alpha'). He warned that relying on 'frontier labs' for AI could lead to these providers competing with their clients by leveraging their data.
Anthropic has been accused of blindsiding business partners, including Figma, with the launch of new vertical applications that compete directly with their partners' offerings. Figma's founder stated that Anthropic was not fully transparent, and the company's stock has seen a significant decline.
The discussion highlighted Anthropic's strategy of launching vertical applications (e.g., Cloud Code, Cloud Design) that directly compete with companies building on their models. This mirrors tactics seen from Microsoft and Google, aiming to dominate the model layer and capture lucrative market segments.
Chamath shared his experience testing open-source AI models, reinforcing the idea that enterprises need to protect their data and intellectual property. He stressed that using third-party AI models without control poses a significant risk of the provider competing with the user's business.
A BCG study indicated that the cost of capital for long-term rates is around 8-11%, yet half of large US companies cannot deliver returns exceeding this. This highlights the difficulty and complexity of running a business, making AI providers a potential risk if they leverage client data to compete.
The discussion emphasized that enterprises can host open-source AI models on their own GPUs and in US data centers, ensuring data security and preventing data sharing with frontier labs. This approach is presented as a more cost-effective and secure alternative to relying on third-party models.
Statistician Nate Silver discusses the overwhelming impact of partisanship on US elections, stating that in 2028, we will likely be able to predict election outcomes in most states with high confidence due to this polarization. He contrasts this with the unpredictability of individual games or player performances.
Nate Silver explains his preference for betting on prediction markets over traditional sportsbooks due to limitations imposed by the latter. He consults for Polymarket and notes that prediction markets, despite their own issues, are more accessible for those considered 'sharp bettors'.
Nate Silver attributes the New York Knicks' defensive success in the latter half of the season and playoffs to coaching system improvements and a potential 'revelation' from Karl-Anthony Towns. He notes that while Jaylen Brunson is a great player, his individual defense is a weak point, but the overall team defense has improved significantly.
Nate Silver stated that there is no evidence of fraud in the California elections discussed, attributing the perceived statistical anomalies to systematic differences in how Democratic voters cast their ballots (later and by mail). However, he criticized California's slow vote-counting process as 'unacceptable' and a 'failed state' system.
One of the speakers describes California as a 'machine state' with strong partisan leanings, contrasting it with New York's historical 'heterodoxy,' which has seen socialist mayors and a wider range of political figures. He notes that expecting a Republican to win a mayoral race in California, even in Los Angeles, is less likely than in New York.
Nate Silver discusses the historical theory that Democrats benefit from increased turnout through any means, contrasting it with a shift where Republicans now also benefit from high turnout. He expresses no opposition to voter ID laws, seeing them as reasonable, but doubts they would explain the observed outcomes in California's elections.
Progressive Democratic Socialists have achieved a significant sweep in New York City's congressional Democratic primaries. Mayor Adams endorsed three candidates who all won their races, indicating a growing influence of the socialist movement within the Democratic Party.
Chamath Palihapitiya stated that Artificial Intelligence has the potential to be the greatest economic leveler, capable of creating equality and leveling the starting line for everyone globally. He contrasted this potential with the perceived mismanagement and negative public perception of AI by Silicon Valley.
Discussing a specific Democratic Socialist (DSA) candidate, the conversation highlighted a platform that includes abolishing the Senate, the carceral state (police and prisons), borders, and the Electoral College. The candidate, described as a 32-year-old PhD candidate who has never held a job, also supports public ownership of corporations and defunding the military.
According to DSA co-chair Josh Block, the Democratic Socialists of America view the Democratic Party primarily as a 'ballot access vehicle.' They aim to get elected on the Democratic label to push their own agenda from the inside, seeing the Democratic establishment as an obstacle rather than an ally.
Jason Sacks predicts a significant defeat for the Democratic Party in the upcoming November elections, describing it as a "shellacking" and a "significant red wave." He believes this outcome is due to traditional Democratic voters, including working-class individuals, immigrants, and minorities, peeling away from the party.