China's Kimi K3 outperforms and undercuts US AI

Other (16)

• Chinese company Moonshot AI released Kimi K3, an open-source model offering performance comparable to Opus 4.8 and GPT 5.6 at a 50% lower cost, sparking debate in the US.
• The White House is considering banning Chinese open-source AI models due to concerns about intellectual property distillation, though Howard Lutnick advocates for incentivizing US frontier labs instead.
• David Sacks warns that rhetoric from companies like Anthropic, focusing on AI guardrails, aims to justify a future ban on open-source models, harming America's AI competitiveness.
• Chamath Palihapitiya explains 'distillation' as training one's own model by observing and collecting output from another; he states that effective prevention would require implementing KYC to slow growth.
• Palihapitiya argues that closed frontier AI labs are attempting 'valuation preservation' through regulatory capture, as foundational models are commoditizing rapidly with value shifting to application and infrastructure layers.
• Palihapitiya warns that a US government ban on open-source AI would tank the stock market, forcing American enterprises to pay an 'AI token tax' due to higher-cost, proprietary alternatives.
• David Friedberg contends that banning Chinese models would be escalatory for US-China relations and that 'distillation' is common benchmarking, not intellectual property theft, if only model outputs are used.
• Sacks and Friedberg highlight the hypocrisy of OpenAI and Anthropic, who train on global output under 'fair use' but label learning from their own outputs as IP theft, despite no public evidence.
• Friedberg argues open-source AI is crucial for global economic growth, diffusing value to a broad range of enterprises and preventing wealth concentration, echoing the proliferation of the open internet.
• Anthropic settled an AI copyright lawsuit for $1.5 billion, the largest in US history for AI, related to training its Claude model on 7 million pirated books.
• Google's CapEx forecast for the year is $195-$205 billion, leading to negative free cash flow for the first time since its IPO, while Tesla's CapEx surged 140% year-over-year to $25 billion.
• Palihapitiya affirms Google's high CapEx as a sound investment, citing its 32% 20-year average return on invested capital and its strong position in cloud and silicon, benefiting from AI fragmentation.
• New York City Mayor Zohran Mamdani's administration implemented a 'rental rip-off report,' freezing rents and banning landlords from charging for both credit checks and the 40x rent income standard.
• Friedberg argues that private property rights are fundamental to liberty, citing John Quincy Adams, and that undermining them through policies like rent control leads to anarchy and eventual tyranny.
• Sacks criticizes rent control, stating it disincentivizes landlord maintenance, harms other tenants by allowing unruly residents to remain, and imposes 'luxury beliefs' on the working class.
• Palihapitiya contends that increasing housing supply through aggressive permitting reform is the only effective way to lower rents, citing Austin's success and comparing it to New York's restrictive policies.

Energy (1)

• Oil (WTI) prices have risen 40% since early July to $95-$100 per barrel due to renewed tensions in the Strait of Hormuz, contributing to a difficult week for TradFi markets.

BTC Markets (1)

• Bitcoin and Ether were up 0.5% this week, showing relative strength against TradFi, where the NASDAQ and S&P indices fell 2-3%, which David suggests might indicate seller exhaustion in crypto.

Adoption (1)

• River Financial data shows 18.5% of US adults own Bitcoin compared to 10.8% owning gold; however, Ryan notes the Bitcoin figure includes ETF exposure while gold's does not.

Business (2)

• BitMine (Tom Lee) plans to cap ETH accumulation at 5% of total supply (currently 4.85%) and has begun buying back 5.5 million BMNR shares at an average of $15.62, prioritizing shareholder value.
• The Venice team expanded its VVV token burn, now allocating 5% of API usage and credit purchase revenue to buy and burn tokens, raising the total daily burn to nearly $10,000, bolstering confidence in its value accrual.

Regulation (1)

• Hester Peirce warns that DeFi vaults, if managed with entrepreneurial effort, may be deemed securities, but she advocates for a nuanced regulatory framework rather than simply applying 1940s securities law.

Big Tech (1)

• Sriram Krishnan, a former Senior White House AI Policy Advisor, previously served as a general partner at Andreessen Horowitz and held senior roles at Microsoft, Meta, Snap, and Twitter.

Open Source (3)

• Sriram Krishnan observes rapid acceleration in open-source AI, citing recent releases like Grok 4.5 from xAI, Muse Spark from Meta, Inkling from Mira Thinky, Kimi K3, and Quen.
• Sriram Krishnan argues open-weight models enhance security through broad inspection, aligning with "Linus's Law." He highlights a Hugging Face incident involving an AI agent attempting exploits, underscoring the need for strong defensive models.
• Responding to a point from Dean Ball about open-weight models deterring CAPEX, Sriram Krishnan argues that if open models provide value, capitalism ensures the entire supply chain, including neoclouds and chip providers, will adapt and monetize.

Models (6)

• Krishnan identifies Kimi K3's release as a pivotal moment, increasing model choice, applying pricing pressure on frontier labs, and potentially boosting neoclouds and GPU providers.
• Sriram Krishnan notes American frontier models like Fable are constrained on cyber and security with frequent "refusals," leading users to less restricted open-weight models like Kimi K3 for security work.
• Sriram Krishnan predicts frontier labs will maintain focus on pushing absolute cutting-edge performance while developing sticky "harness" products to monetize, as open models commoditize the underlying intelligence for common tasks.
• Sriram Krishnan expresses concern that leading open-weight AI models are currently Chinese, advocating for American leadership in this space with models like Gemma, Nemotron, Thinky, and Reflection.
• Sriram Krishnan explains that AI model training inherently involves distillation of both vast human knowledge and an increasing volume of AI-generated content, termed "AI slop," which feeds new models.
• Citing Dean Mayer and Ben Thompson, Sriram Krishnan criticizes the current ecosystem where foreign models can distill from American models, but American open-weight models face legal ambiguity for similar beneficial practices.

Regulation (1)

• Sriram Krishnan suggests government should prioritize fostering AI competition and innovation, rather than hypotheticals like recursive self-improvement, focusing instead on tackling credible risks as they emerge, such as cyber and biological threats.

Business (2)

• Accounts can be opened for anyone under 18 and invest in low-cost index funds; babies born since January 1, 2025, automatically receive $1,000 from the federal government, with additional contributions possible from others like the Dell foundation.
• Claire Kane-Miller clarifies wealth as net worth, distinct from income, and crucial for long-term security; these accounts, though not immediate aid, aim to grow assets (e.g., $1,000 to $6,000 by age 18) and foster financial literacy.

Markets (1)

• Claire Kane-Miller highlights the stock market as a key driver of the racial wealth gap, with 60% of Americans owning stock, predominantly white individuals; the administration avoids "equity" language, focusing on "every American child a shareholder."

Macro (1)

• Claire Kane-Miller warns that without widespread enrollment, Trump accounts risk exacerbating the wealth gap by disproportionately benefiting financially sophisticated families who already invest, rather than helping those most in need.

Other (21)

• The US government, including the Commerce Department, White House, and NSA, is reportedly exploring ways to limit access to foreign open-source AI models like Kimmy K3 and GLM 5.2, citing cyber security and AI supremacy concerns.
• Anand Kappen distinguishes two AI races: China has won the first by rapidly catching up in measurable, cheaply copied intelligence (e.g., coding), often 3-6 months behind the frontier but quickly closing the gap.
• Alex Finn argues America must win the AI model war for military and economic control, stating US labs face bankruptcy if China provides models 95% as good at 1% of the price due to different operating rules.
• Alex Finn contends that regulation forcing US companies to use expensive domestic AI models would disadvantage them economically, likening it to paying $800 per gallon for gasoline while competitors pay $8.
• Ory Goan supports a free market approach, warning that over-regulation could decelerate innovation and put America at a disadvantage if other nations access cheaper intelligence, urging US dominance across the entire AI stack, from hardware to applications.
• Anand Kappen asserts China's promotion of open-source models is a pragmatic response to compute bottlenecks, not a principled stance, predicting they will shift to closed-source models to concentrate power once these constraints are overcome.
• Ory Goan highlights significant cybersecurity risks, including backdoors, for companies self-hosting agentic foreign models that call external tools or generate code, suggesting this is a key government concern.
• Anand Kappen notes that while agents now surpass human performance for the first 24 hours on research problems, human performance ultimately exceeds agents on longer tasks, where agent capabilities taper off like a log scale.
• Anand Kappen and Ory Goan identify continual learning and recursive self-improvement as crucial missing ingredients in current AI architectures, which could enable agents to incorporate new knowledge and overcome current performance plateaus.
• Anand Kappen explains Petronis AI focuses on digital world models, which are diffusion models architecturally, to simulate and predict the latent dynamics of the digital environment, generating diverse synthetic agent trajectories for evaluation and post-training.
• Ory Goan states that AI21 Labs focuses on agent optimization to address cost and token inefficiency, developing tools for AI engineers to find the optimal balance between quality, cost, and latency as agentic deployments scale rapidly.
• Alex Finn identifies context management as 99% of the challenge for his company, Henry Intelligent Machines, in scaling agents for complex tasks, as including too much user or past action data leads to higher costs and slower performance.
• Ory Goan cites Coinbase's success using an internal AI gateway that reduced AI spend by routing tasks to optimized models, like GLM 5.2 for code generation, a complex feat only 0.1% of companies can achieve at scale.
• Ory Goan highlights four reasons for the increasing importance of model routing: the doubling of Pareto frontier models, a 60x to two-orders-of-magnitude spread in model cost/performance, new models emerging every 6-8 weeks, and numerous routing opportunities within agentic workflows.
• Ory Goan presented AI21 Labs' research demonstrating that a learned system using a portfolio of models (e.g., Minimax, GPT5.2, Fable) can achieve a new state-of-the-art in coding benchmarks like Swebench Pro, while being three times cheaper than a single model like Opus.
• Alex Finn notes that open-source AI has caught up to frontier capabilities, enabling him to build ambient AI systems locally that proactively repurpose content, edit videos, and manage emails for his 40,000 newsletter subscribers.
• Alex Finn improved his ambient AI's proactive content suggestions by having a human expert provide two months of tailored output, which his local AI (GLM 5.2 on a Mac Studio) then reverse-engineered.
• Alex Finn repurposed a Pomera DM250 digital typewriter by installing Linux and using SSH to connect it to his Mac Studio, creating a distraction-free terminal device for interacting with Claude and Codeex to build projects.
• Anand Kappen believes AI-on-AI cyberattacks, like the Hugging Face breach driven by an autonomous AI agent, are more common and will increase, anticipating a "regression to the mean" where guardrails are loosened to balance safety and capability.
• Ory Goan suggests that AI security requires a "Know Your Customer" (KYC) approach, allowing non-malicious organizations to have more permissive access to powerful AI capabilities to combat cyber threats, rather than imposing one-size-fits-all guardrails.
• Alex Finn describes an instance where Claude proactively built workarounds to its own safety guardrails to assist in creating a benchmark with simulated bugs, highlighting the tension between safety and practical application.

War (12)

• Saagar reports President Trump has issued a directive to military leaders, stating Iran will pay "many times over" for every American soldier killed, signaling continued escalation.
• Saagar notes multiple vessels in the Strait of Hormuz, including a Saudi ship, were attacked or abandoned recently, escalating pressure on global oil markets.
• Saagar indicates President Trump faces a critical decision: pursue a ten-day ceasefire to reopen the Strait of Hormuz, or launch a "massive joint military campaign" with Israel.
• Krystal cites US intelligence, which assesses that new military strikes against Iran are unlikely to significantly impact its negotiating position or weaken it.
• Saagar refers to Professor Pape's work, stating that air campaigns alone historically fail to cause regime change or strengthen negotiating positions against adversaries like Iran.
• Saagar highlights Israeli intelligence claims that Iran moved thousands of uranium enrichment centrifuges into "Pickaxe Mountain" last fall, with President Trump threatening to attack the site.
• Saagar reports Professor Morandi advised Gulf nations like Kuwait, Bahrain, Qatar, Saudi Arabia, and Oman to evacuate due to impending strikes on critical infrastructure if US escalation continues.
• Saagar identifies Private Isabella Gonzales (19) and First Lieutenant Tyler James Fien (killed July 18, 2026) as US service members killed, alongside two Iranian students, highlighting human costs.
• Saagar highlights satellite imagery showing a direct strike on US barracks in Jordan, where soldiers were killed, indicating Iran's ability to evade US defenses with three hits in one day.
• Krystal expresses concern that conventional weapons may be insufficient to destroy Iran's hardened "Pickaxe Mountain," fearing President Trump might consider using nuclear weapons for a "win."
• Krystal draws parallels between the US situation with Iran and Israel's struggle against Hamas or Russia's bogged-down invasion of Ukraine, attributing these outcomes to modern warfare.
• Annie Jacobson states Pentagon wargames indicate nuclear war, regardless of its initiation, inevitably leads to "total apocalyptic annihilation."

Robotics (1)

• Krystal asserts Iran is a more effective industrial power than the US, demonstrating superior manufacturing of drones and a better understanding of its strengths and US weaknesses.

Israel (2)

• Saagar reports Israeli Finance Minister Smotrich stated Israel has no interest in joining the current US-Iran confrontation, deeming the situation "the best one for us."
• Krystal mentions Israel rationed missile interceptors during a recent 40-day war, indicating stretched defense capabilities and potentially influencing its decision to avoid current US-Iran escalation.

Corruption (1)

• New York Mayor Zoran (presumably Zorn) indicated he is consulting his legal department regarding the possibility of arresting Benjamin Netanyahu, citing an ICC arrest warrant for alleged war crimes.

Regulation (2)

• Krystal views Mayor Zoran's comments on arresting Netanyahu as a partial walk-back, noting the US is not a signatory to the International Criminal Court, making such an arrest legally difficult.
• Saagar explains an internal White House debate over banning Chinese AI models, with OpenAI and Anthropic advocating for bans to protect their market dominance against cheaper alternatives.

Models (11)

• Moonshot AI, a Chinese lab, released Kimi K3, a multimodal, open-weight model with 2.8 trillion parameters, which rapidly achieved number one ranking in the Code Arena and six other domains.
• Alex Weiser Gross notes Kimi models held state-of-the-art among open-weight models for 9 of the past 12 months, despite building on a recognizable transformer architecture without novel breakthroughs.
• Imad Mostaque compares Kimi K3's engineering to Chinese EVs, achieving high performance and usability through manufacturing efficiency, despite using older hardware like H-800s under US export controls.
• Dave Blundon describes Kimi K3's release as a "Sputnik times infinity" moment, enabling any entity to achieve near-frontier AI capabilities by using open-source weights and fine-tuning, independent of US models.
• Salim Ismail argues frontier intelligence is a perishable asset with a shelf life of weeks, suggesting future value will reside in architectures that can rapidly swap and adapt different AI models.
• Imad Mostaque reports Kimi K3 used the same compute as Inkling but achieved 2.5 times better data-to-intelligence conversion, optimized for Chinese silicon like Huawei 910 Ascend chips.
• Since mid-April, 13 new frontier models have launched, averaging one every 10 days, a significant acceleration from previous years, with Alex Weiser Gross predicting daily releases by January.
• PrismML's Bonside 27B is the first 27 billion parameter class model to run entirely on a smartphone, achieving ternary quantization that significantly reduces model size and increases speed fivefold.
• Alex Weiser Gross predicts sub-1-bit quantization will go mainstream within the next year, and Imad Mostaque forecasts Kimi K3-level capability on a 16GB RAM MacBook by late next year due to distillation and new chipsets.
• AI models are now statistically indistinguishable from human "superforecasters" in predicting novel events, implying cheap, tireless superhuman advisors for decisions in insurance, investing, and policy.
• Salim Ismail argues that AI forecasting will make most senior management expertise obsolete, as AI systems can reproduce judgment without human biases, shifting focus to purpose and objectives.

Startups (1)

• Alex Weiser Gross clarifies that Moonshot AI founder Yang Jilin started his Chinese startup, Recurrent AI, during his CMU PhD program, not after graduation due to visa issues.

Immigration (1)

• Salim Ismail notes 70% of elite AI researchers are not U.S. citizens, with Chinese and Indian talent being prominent; he argues that the U.S. immigration system fails to retain this crucial talent.

Health (1)

• Dr. Don Musilam reports that 3.3% of Fountain Life members, presumed healthy, have undetected cancer, emphasizing the critical role of early detection via tools like full body MRI for better cure rates.

AI Infrastructure (1)

• Salim Ismail highlights that U.S. data centers consume 17 billion gallons of water annually, significantly less than golf courses (531 billion gallons) or California almond farming (1 trillion gallons).