Andrej Karpathy — AGI is still a decade away

Andrej Karpathy — AGI is still a decade away – Dwarkesh Podcast

Mentioned

Bullish

AIEDU— AI will fundamentally transform education delivery through personalized tutoring and efficient learning ramps, making education both more useful pre-AGI and more engaging post-AGI
CODINGAI— Coding is the perfect first application for LLMs and agents due to its text-based nature and existing infrastructure for automation
CALLAI— Call centers have ideal properties for near-term AI automation due to their simple, repetitive, and similar task sequences
SEMI— Semiconductor demand will absorb current infrastructure buildout as compute requirements continue to grow
TSLA— Tesla's self-driving approach is more scalable than competitors despite current market perceptions

Bearish

AGITIMELINE— AGI timeline predictions are overhyped and driven by fundraising rather than technical reality, with no historical precedent for discrete intelligence explosions
AGENTS— AI agent development will take much longer than market expects, with the industry making overpredictions about near-term capabilities
WAYMO— Waymo's business model is uneconomical with very few cars deployed due to fundamental economic challenges
CODEAGENT— Production-grade coding agents face significant challenges due to safety and reliability requirements that current systems cannot meet
SLIDEAI— Presentation and slide automation is much harder than text-based AI due to lack of prebuilt infrastructure and visual complexity
RADAI— Radiology AI predictions have proven wrong with radiologists remaining alive, well, and growing despite automation predictions

Not investment advice.