newsfilter.io
Interview, Fireside Chat

AI Fund’s GP, Andrew Ng: LLMs as the Next Geopolitical Weapon & Do Margins Still Matter in AI?

  • Demand for AI token generation is expected to remain insatiable in the short term due to constraints in semiconductors, data centers, and electricity, preventing supply from meeting demand despite increased efficiency and lower costs.
  • The industry anticipates a correction of the current VC-subsidized application era, driven by financial and physical laws, leading to a future where valuable businesses operate without perpetual external capital.
  • AI-assisted coding is predicted to deliver massive productivity gains, reducing tasks that previously required six engineers over six months to one engineer in a weekend, making a return to manual coding unlikely.
  • Defensibility for businesses is expected to shift away from software code due to reduced copying time, favoring alternative moats such as brand reputation, industry-specific data, two-sided marketplace dynamics, or ecosystem integration.
  • The transition from human labor budgets to software budgets is viewed as crucial for unlocking massive GDP growth, with the industry expected to move from automating individual steps to rethinking entire workflows.
  • Geopolitical dynamics will be shaped by open-weight models, with China predicted to leverage state-level investments in semiconductors and education to project soft power and build a commanding user base globally.
  • US export controls on semiconductors are expected to have backfired by incentivizing China to accelerate its own semiconductor development, potentially to US long-term disadvantage.
  • Large enterprise AI adoption is predicted to proceed much slower than market hype suggests, with many organizations remaining on-premises or resisting third-party tools in the near term.
  • Token prices are expected to continue falling rapidly, with companies able to bend the cost curve down faster than prices drop, despite current high API costs.
  • The software industry is forecast to evolve from vertical integration to horizontal specialization as standards mature, allowing players to build compatible components similar to the shift from IBM-era integration to USB standards.
  • A dilemma for venture capitalists is anticipated as the cost to try new application-layer ideas remains extremely low (around $1 million), making it difficult to deploy massive capital effectively.
  • A skills gap is predicted where fresh college graduates lacking AI proficiency will struggle, while experienced engineers adept at AI tools will become significantly faster than previously seen.
  • Useful AI agentic workflows for complex tasks such as tariff compliance, medical assistance, and legal processing are expected to exist today rather than being a decade away.
  • Hype surrounding immediate human extinction or near-term AGI is expected to distort public perception and discourage high school students from pursuing AI careers, despite such scenarios being decades away.
  • Intelligence is predicted to become cheap, allowing individuals to access "armies of smart, well-informed staff" for medical, educational, and financial advice, thereby democratizing services previously available only to the wealthy.
  • The market will likely feature a coexistence of large monolithic models and tiny small models to handle a huge range of tasks, mirroring how humans perform tasks of varying difficulty.
  • Large monolithic models are not expected to dominate all use cases, and the "software mode" of defensibility is predicted to have weakened significantly due to reduced time to copy.
  • The transition to AI coding will allow developers to generate code for simple personal projects quickly, demonstrating low economic value tasks that can be completed rapidly.
  • Recruiters are expected to screen significantly more resumes using AI, reducing the need for multiple analysts, while other job functions like marketing and finance may evolve similarly.
  • AI adoption in large enterprises will eventually enable decision-making times to shift from two weeks to 10 minutes, changing products and driving growth rather than just reducing costs.
  • Europe is predicted to have lost ground in the AI race due to over-regulation, with suggestions that regaining equality requires stopping regulation to focus on investment and building.
  • The bipartisan Schumer AI Insight Forum is expected to have successfully mitigated stifling or anti-competitive regulations aimed at shutting down open source, positioning the US to invest in AI.
  • High school students are expected to be guided to learn to code and understand how to direct computers to perform tasks, which will be more powerful and fun than writing code by hand.
  • Data will remain a verticalized asset, allowing businesses to build valuable applications using stripped internal or public data rather than requiring massive general-purpose internet data.
  • Circular deals and complex financial instruments in the AI sector are viewed as signs of frothiness but are not expected to necessitate alarm, with margins expected to matter eventually.
  • China's government is expected to continue making all industrial commitments to AI, leveraging the state apparatus to control rare earth elements and drive velocity.
  • Verticals in AI such as travel, retail, and transportation are expected to emerge around dominant horizontal platforms, creating opportunities for numerous vertical applications.
  • The industry will likely not replace the bottom 5% of workforce capabilities as a primary success metric, but rather focus on 10x-ing human ability.
  • AI is expected to replace many junior ranks in fields like consulting and law, potentially creating a talent pipeline issue, though this problem may be less severe than predicted.
  • AI will allow businesses to serve larger groups with high-touch services like financial advice, thereby lowering economic barriers and driving growth.
  • Medical discoveries are expected to accelerate in diseases with little prior advancement, such as MS, driven by AI capabilities.
  • AI will allow developers to get more done than those who cannot code, even for simple tasks, by enabling them to tell the computer exactly what they want done.
  • Smaller applications are predicted to continue growing as they are not yet generating hundreds of millions in revenue but are inexpensive to build and operate.
  • AI will not make humans lazy or reduce efficiency despite wealth increases, as observed in high-performing Silicon Valley individuals.
  • Only a small subset of jobs will see headcount reductions from AI screening, while the vast majority of knowledge workers will retain roles AI cannot perform.
  • AI will not be able to do everything a human can do for decades, with 50% to 70% of tasks still requiring human intervention.
  • The era of "all-nation commitment" to AI in China is characterized by a powerful force of state investment that will not be underestimated.
  • The current era of VC-subsidized AI applications is similar to early food delivery and is expected to correct as laws of physics or finance dictate.
  • By 2030 or within the next decade, the industry will still be actively identifying valuable applications and building them, continuing beyond the current one or two years.
  • The speaker predicts that the "software mode" of defensibility has weakened significantly because the time to copy is reduced significantly, requiring new moats.
  • AI adoption will lead to tremendous GDP growth, but the timeline for full implementation will be much longer than the hype suggests.
  • AI will enable a shift from "do more" or "do it faster" to drive growth, such as turning loan approval times from two weeks to 10 minutes.
  • Open-weight models from China will increase geopolitical influence by building a commanding user base in the global AI supply chain.
  • Hype from companies facing existential risk will be worse and often weird or desperate, whereas established companies will tend to say more sensible things.
  • AI coding assistants will serve as a foreshadowing for how other job functions will evolve as tools for those sectors become more efficient.
  • Universities will need to ensure CS majors know how to use AI, as the current cohort without AI skills is struggling.
  • Demand for AI coding assistance will be through the roof due to increased developer productivity and efficiency.
  • AI will blend into a 5, 6, or more percent GDP growth scenario, making intelligence cheap.
  • Verticals in AI will emerge around dominant horizontal platforms, creating room for lots of verticals to be built out.
  • AI will not replace the bottom 5% of capabilities as a primary success metric, but rather the real question is whether it can 10x people's ability.
  • AI will replace many junior ranks in fields like consulting and law, potentially creating a talent hole.
  • AI will allow businesses to turn around decision-making time from two weeks to 10 minutes, thereby changing the product.
  • AI will allow financial advice to be delivered to a lot more people, changing the product.
  • AI will allow developers to get more done than those who can't code.
  • AI will lead to smaller applications that are not yet doing hundreds of millions in revenue but will continue to grow.
  • AI will allow businesses to serve a much larger group of people by lowering the economic barrier.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards for personal projects.
  • AI will allow recruiters to screen a lot more resumes using AI.
  • AI will not make humans lazy or lose efficiency.
  • AI will not be able to do everything a human can do for decades.
  • AI will allow businesses to rework workflows to turn around decision-making time.
  • AI will allow financial advice to be delivered to a lot more people.
  • AI will allow developers to get more done.
  • AI will allow AI-assisted coding to be used to write code for generating flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.
  • AI will not make humans lazy.
  • AI will allow developers to get more done.
  • AI will lead to smaller applications.
  • AI will allow businesses to serve a much larger group of people.
  • AI will allow AI-assisted coding to be used to generate and print out flash cards.
  • AI will allow recruiters to screen a lot more resumes.