Pessimistic Views on AI Development and Their Investment Implications During these days of researching AI, Sam has encountered mostly optimistic perspectives. But occasionally, he came across pessimistic, opposing, and skeptical viewpoints. Sam found them fascinating, so he summarized them as follows: The Pessimists: Demis Hassabis (CEO of Google DeepMind) Insight: The concept of "jagged intelligence": "AI can achieve Olympic gold-medal-level mathematical ability, yet fail at problems a 12-year-old child could solve." Ilya Sutskever (Founder of Safe Superintelligence Inc.) Core View: We have returned from the "scaling era" to the "research era." The revolutionary nature of this view: Over the past five years, the entire AI industry—including NVIDIA’s trillion-dollar orders—was built on the assumption that "if we just keep scaling, AGI will emerge." Sutskever—the very person who first bet on the scaling approach—is now openly declaring, "The scaling era is over." This forms a key basis for Sequoia Capital’s assertion that "2026 is the year of delay." Andrej Karpathy (Former Director of AI at Tesla, Founder of Eureka Labs) Core View: AGI is still 10 years away; current AI is "slop." Karpathy’s statements over the past six months have been the most blunt among all industry leaders. His exact words: "I think the industry is making too big a leap, pretending everything is astonishing when it’s not. It’s slop." Richard Sutton (2024 Turing Award winner, "Father of Reinforcement Learning") Core View: LLMs are fundamentally not intelligence—they merely mimic intelligent things (humans). The Opponents: Yann LeCun (Formerly at Meta, now founded AMI Labs) We’ve discussed this in detail before. His core view: The LLM architecture is fundamentally flawed; we must shift to the JEPA (Joint Embedding Predictive Architecture) pathway. Fei-Fei Li (Founder of World Labs) Core View: AGI without spatial intelligence is incomplete. Geoffrey Hinton ("Godfather of AI," 2024 Nobel Prize in Physics winner) Core View: AI is developing faster than I imagined; by 2026, "many, many" jobs will be replaced. Hinton’s stance is unique: He is a proponent of scaling (believes continued scaling will work), yet also a doomsayer (warns of existential risks). The Skeptics: Gary Marcus (Renowned cognitive scientist) Core View: The AI industry has reached its "bubble peak"; the illusion of AGI is about to collapse. Daron Acemoglu (MIT economist, Nobel laureate in Economics) Core View: AI’s economic contribution has been grossly exaggerated. "Acemoglu predicts AI will contribute only 1.1%–1.6% to GDP growth over the next decade. He estimates AI will automate only 5% of work tasks. 'Hype is the enemy of business success.'" Sequoia Capital’s Position In their late 2025 article, “AI in 2026: A Tale of Two AIs,” Sequoia stated: “For years, Silicon Valley luminaries have predicted AGI is imminent.” Since June this year, that timeline has been consistently pushed back. Dwarkesh Patel’s recent podcast interviews with Sutton, Karpathy, and Sutskever marked a turning point; the new consensus is that the AGI window opens no earlier than the 2030s. Investment Implications: 1. It is currently unwise to heavily bet on the rapid arrival of AGI—it is likely to be repeatedly delayed. 2. Once the industry collectively acknowledges that scaling LLMs alone cannot achieve AGI, valuations across the LLM supply chain will experience a sharp correction. 3. Therefore, it may be prudent to allocate some capital toward alternative model architectures beyond LLMs. P.S. Sam himself remains optimistic about AI’s development—and he must remain so, because humanity has no other technological红利 left.