Marc Andreessen and Ben Horowitz on the State of AI
- By
- a16z
- Length
- 39 min
- Language
- English
- Text from
- the video's captions
- Transcribed
- 14 October 2026
Summary
Marc Andreessen and Ben Horowitz, joined by a moderator, discuss the state of AI at an a16z event. They debate whether LLMs can be truly intelligent and creative, arguing that most humans rarely clear those bars either, and explore why raw IQ isn't sufficient for leadership. They also cover the AI bubble question, incumbent versus startup dynamics, and the US-China AI race, with Marc warning that phase two of the AI revolution will be robotics, where China's industrial ecosystem gives it an edge.
Key points
- Marc Andreessen argues that if LLMs clear the bar of 99.99% of human intelligence and creativity, that is significant in itself, since most people never have true conceptual breakthroughs.
- Ben Horowitz notes that hip-hop artists are especially interested in AI because sampling and remixing mirrors how the genre was built.
- Marc cites the US military's finding that leaders more than one standard deviation of IQ away from their followers lose theory of mind in both directions.
- Ben argues the current AI moment is not a bubble because demand is real and customers are paying, and a bubble requires everyone believing it isn't one.
- Marc compares today's chatbots to the text-prompt era of personal computers, predicting radically new user experiences will emerge.
- Marc highlights strong Chinese models like DeepSeek, Qwen and Kimi, built by teams that are not the name-brand researchers.
- Marc warns that phase two of the AI revolution is robotics, and China's industrial ecosystem gives it a potential hardware advantage.
- Ben advises entrepreneurs to think from first principles because this era's companies and talent are genuinely different from prior generations.
Questions it answers
01:00LLM intelligence and creativity
Can LLMs truly invent or create, or are they just remixing?
Marc argues most people never have original conceptual breakthroughs either, and even historic breakthroughs build on decades of prior work, so if models clear the bar of 99.99% of humanity that is already remarkable.
- Marc knows only about three people out of 10,000 contacts who reliably reason outside distribution.
- Language models are the culmination of eight decades of prior work.
05:00Hip-hop and creative genius
How do creative artists view AI?
Ben says hip-hop artists are very interested in AI because sampling and remixing mirrors how the genre was built, and AI opens up the creative palette.
- Ben names Rakim, Dr. Dre and George Clinton as true conceptual innovators.
- Hip-hop's specific time-and-place storytelling favors models trained on that context.
09:00Intelligence vs. other factors
Why isn't raw intelligence enough for leadership?
Marc notes IQ correlates only about 0.4 with outcomes, and military research shows leaders more than one standard deviation of IQ from followers lose theory of mind in both directions.
- Ben says leadership involves confrontation, theory of mind, and getting people to do what's correct, not popular.
- A machine with a 1000 IQ might be too alien to connect with the people it manages.
19:00Theory of mind in current models
How good are today's models at modeling people?
Marc says current models are very good at theory of mind — he uses them for Socratic dialogues with personas, and a UK startup found models can accurately simulate voter focus groups.
- Models default to making personas agree; prompting for tension makes dialogues more interesting.
- Simulated focus groups could replace expensive, slow physical ones.
23:00Is AI a bubble?
Is the massive AI infrastructure buildout a bubble?
Ben says a bubble requires capitulation and everyone believing it isn't a bubble, whereas AI has real short-term demand and customers paying; Marc says to check two fundamentals: does the technology work, and are customers paying.
- The internet wasn't a bubble long-term despite short-term price dislocation.
- Current supply and demand dynamics don't look like a bubble to Ben.
27:00Incumbents vs. new entrants
Do incumbents like Google win the AI platform shift?
Ben says new companies have historically won new markets while incumbents persist on prior monopolies; Marc adds that the ultimate product forms are still unformed, like the PC's shift from text prompts to GUIs to browsers.
- Google woke up after ChatGPT, but OpenAI isn't going away.
- There will likely be chatbots in 20 years, but also radically different user experiences.
31:00Advice for entrepreneurs
What should entrepreneurs do differently in this era?
Ben says this era is genuinely different, so entrepreneurs should think from first principles rather than copying past organizational lessons; Marc predicts current shortages of researchers and chips will eventually become gluts.
- Chinese teams like DeepSeek, Qwen and Kimi are producing excellent models without name-brand researchers.
- Every chip shortage in history has led to a glut.
35:00China and robotics
How should we think about the US-China AI race?
Marc says China is running the catch-up game very well and the US lead may be six months, not five years; he warns phase two is robotics, where China's industrial ecosystem could let it lap the US in hardware.
- The US shouldn't constrain its companies in ways China doesn't.
- Robotics requires an entire ecosystem of component suppliers, which China has today.
- Marc is guardedly optimistic about growing US awareness that de-industrialization went too far.
Notes
Intelligence and creativity of LLMs
- Marc Andreessen responds to claims that LLMs can't do true invention by asking: can people? He says he has met only a few people capable of genuine conceptual breakthroughs, and true creative geniuses like Beethoven or Van Gogh are rare.
- He notes that big breakthroughs in technology and art typically build on decades of prior work — language models themselves are the culmination of eight decades of research — so "remixing" is the norm even for humans.
Hip-hop and AI in music
- Ben Horowitz, whose Paid in Full Foundation supports hip-hop innovators, names Rakim, Dr. Dre and George Clinton as true conceptual innovators in the field.
- Hip-hop artists are especially interested in AI because the genre was built from sampling and remixing other music, making AI a natural creative tool.
Intelligence isn't everything
- Marc points out that IQ correlates about 0.4 with life outcomes — large for social science, but far from determinative.
- The US military found that leaders more than one standard deviation of IQ away from followers lose theory of mind in both directions.
- Ben adds that leadership requires confrontation skills, understanding what people want, and getting people to do what's correct rather than popular — situational skills AI could in principle learn but nobody is currently training on.
- Marc suggests human cognition is a full-body experience, and today's disembodied AI will need robotics to integrate physical experience.
Theory of mind in today's models
- Marc says current models are very good at theory of mind; he uses them to create Socratic dialogues with personas, though he has to prompt them to add tension and conflict.
- A UK startup in politics found models can accurately simulate focus groups of real voters.
Bubble question
- Ben argues this isn't a bubble: a bubble requires capitulation and everyone believing it isn't one, while AI has real short-term demand and customers paying.
- Marc says nobody knows, but ground-truth fundamentals are: does the technology work, and are customers paying for it?
Incumbents vs. new entrants
- Ben says Google woke up after ChatGPT, but historically new companies win new markets while incumbents persist on prior monopolies.
- Marc compares chatbots to the text-prompt PC era (1975–1992) before GUIs and browsers changed everything — the ultimate product forms are still unknown.
China and robotics
- Marc says China (DeepSeek, Qwen, Kimi) is running the catch-up game very well, and the US lead may be six months, not five years.
- He warns phase two of AI is robotics, where China's industrial ecosystem could let it lap the US in hardware despite US software strength.
Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.
More from a16z
- Marc Andreessen: How Movies Explain America
YouTube · 1 h 16 min
- Ben & Marc: Why Everything Is About to Get 10x Bigger
YouTube · 55 min
- Sacks, Andreessen & Horowitz: How America Wins the AI Race Against China
YouTube · 1 h 17 min
- Marc Andreessen's 2026 Outlook: AI Timelines, US vs. China, and The Price of AI
YouTube · 1 h 21 min