Best of the Pod: Reid Hoffman on How AI Is Answering Our Biggest Questions

Hoffman explains why he tells MBA students that philosophy is more important than business school for entrepreneurship. Philosophy trains you to think crisply about possibilities, theories of human nature, and how circumstances change—all critical for building products and services that shape human behavior.

• •Philosophy teaches you to think about what's possible and theories of human nature
• •Entrepreneurs must understand how technology modifies human behavior and creates new possibilities
• •Philosophy helps you develop working theories that evolve dynamically, similar to science
• •The discipline provides frameworks for thinking about unanswerable questions that are actually central to innovation

Hoffman critiques how thought experiments like trolley problems are misused in philosophy and EA discourse. He argues they artificially constrain options by positing perfect knowledge in unrealistic scenarios, when the correct response might be to reject the framing entirely and attempt a third option.

• •Trolley problems frame artificially different environments to drive intuitions about principles
• •They often posit perfect knowledge that we never encounter in reality
• •The right response may be to reject the binary choice and try to 'break the trolley'
• •This critique applies to certain doom arguments that hijack instincts with hidden assumptions

Hoffman explores how LLMs function more like later Wittgenstein's language games than essentialist approaches to truth. While they're built on pragmatic, nominalist foundations (next token prediction), we're now trying to ground them in more essentialist characteristics like truth and reduced hallucination—creating a Hegelian synthesis.

• •LLMs operate through language patterns similar to Wittgenstein's 'language games' and forms of life
• •Early AI approaches that tried to list all objects failed; pragmatic token prediction succeeded
• •We're now adding essentialist grounding (truth, reduced hallucination) to the nominalist base
• •Truth is a dynamic process in human discourse, not just what's expressed in language

The discussion examines how word embeddings in high-dimensional spaces relate to early Wittgenstein's logical possibility spaces. While embeddings seem to map words into atomic facts, their construction is actually late Wittgensteinian—based on usage in practice rather than pure logical ordering.

• •Embeddings represent words in high-dimensional spaces with context-dependent meanings
• •This resembles early Wittgenstein's atomic facts and logical possibility spaces
• •However, the space is constructed based on actual language use, making it late Wittgensteinian
• •Early Wittgenstein thought correct logic would prevent truth mistakes; LLMs prove otherwise

Hoffman outlines current approaches to improving LLM reasoning: training on computer code (which teaches crisp reasoning patterns), using textbooks (human-designed reasoning curricula), and potentially incorporating philosophy of science frameworks. The challenge is that coherent language use can still produce falsities.

• •Training on computer code helps models learn reasoning patterns beyond just coding
• •Textbooks provide structured reasoning curricula that work for smaller, effective models
• •Philosophy of science (Popper, Kuhn, Lakatos) could inform better reasoning architectures
• •Synthetic data generation may benefit from philosophical frameworks about reasoning patterns

Hoffman discusses Gödel, Escher, Bach and how Gödel's incompleteness theorem—that any robust language system contains truths it cannot express—relates to LLM development. He speculates on what a conversation between Gödel and Wittgenstein about truth and logic would reveal about AI systems.

• •Gödel proved that sufficiently robust language systems contain inexpressible truths
• •This recursion pattern appears in Escher's art and Bach's music
• •LLMs face similar boundaries between what can be enumerated and what exists beyond enumeration
• •The dynamic synthesis process (Hegelian thesis-antithesis-synthesis) is key to truth discovery

Hoffman argues that humans are not static but constituted by the technology we use—from glasses to language models. Technology changes our perception, which fundamentally shapes our theories of truth. This cultural evolution is much faster than biological evolution and represents the 'secret of our success' as a species.

• •Technology like glasses fundamentally changes how we perceive and understand truth
• •We evolved from geological to biological to cultural/digital evolution (much faster)
• •Reading literally changes brain structure and creates different psychology (per Joseph Henrich)
• •LLMs are tools to accelerate cultural-digital evolution, like personal research assistants

Hoffman critiques academic philosophy's disciplinarianism and argues that philosophers' focus on pure cogitation prevented them from engaging with technology as a way of knowing. Computer scientists, who asked 'what can I make with this technology?', were better positioned to create AI.

• •Academic disciplinarianism limits cross-pollination between philosophy and technology
• •Philosophers traditionally reject technology as necessary for reasoning (Cartesian cogitation)
• •Computer scientists approached AI by asking what could be constructed with technology
• •Philosophy should be broader and more engaged with the world, not just trolley problems

Hoffman shares his practical approach to using ChatGPT for philosophical thinking: presenting arguments and asking for both supporting arguments and counterarguments to achieve synthesis. He recommends using it to understand complex ideas like Gödel's theorem and engaging dynamically with canonical works.

• •Use ChatGPT to generate arguments for and against your positions (thesis-antithesis-synthesis)
• •Ask it to explain complex ideas (Gödel's theorem, Einstein's relativity) at your level
• •Dynamic engagement with ideas is more important than any single book
• •Read canonical works like Gödel, Escher, Bach to see amazing thinking in action