The 'AGI Is Here' Debate: From Investor Claim to Schmidhuber's Rebuttal

Episode 1 · Investor Claims AGI Has Already Arrived (2026-09-07, 2 posts)

Investor craigweiss declared AGI has already arrived, sparking debate; respondents agreed models rarely get stuck anymore but remain skeptical about original breakthroughs.

• craigweiss claims it's "pretty fair to say that AGI is here" — craigweiss · 2026-09-07
• Reply to "AGI is here": models rarely stall now, novel breakthroughs still lacking — gandamu_ml · 2026-09-07

Episode 2 · Schmidhuber Rejects "AGI Is Here": No Physical-World Mastery, No AGI (2026-09-07, 8 posts)

On September 8, deep learning veteran Jürgen Schmidhuber publicly dismissed claims that "AGI is here" — including investor Chamath's remark that "AGI has arrived and the next 18 months will be crazy" and similar statements by Jensen Huang — calling them absurd. His core criteria: no AGI without mastery of the physical world, and no genuine self-improvement without self-improving hardware. His verdict: AGI has not arrived.

Confirmed

• He argues that today's genuinely useful AI exists only in the virtual world behind screens (summaries, images, programs, slides); no AI robot can do a plumber's job, or even match a capuchin monkey.
• In an interview he reiterated that LLMs like ChatGPT are far from AGI — essentially a clever way of indexing existing human knowledge in natural language, capable of desk work such as summarizing and illustrating, and thus able to replace many desk jobs, but nothing more.
• On self-improvement: current self-improvement is limited to software-level meta-learning; genuine self-improvement requires self-improving hardware.
• Citing Section 20 of his long survey DLH, he notes that humans have imagined physical self-reproducing machines (SRMs) for centuries — from Descartes to von Neumann to Zuse — without knowing how to build one; AI must control the physical world through physically self-reproducing robots, and an AI that builds robots may be the first practically realizable SRM — his answer to what is still missing for true AGI.

Why it matters

• The dispute goes to the definition of AGI: whether "screen intelligence" suffices or embodiment in the physical world is required. Schmidhuber offers operational technical criteria (physically self-reproducing robots, self-improving hardware) as a stricter standard for assessing AGI timelines.
• He acknowledges LLMs' capacity to replace desk jobs, meaning the economic impact is real even if, by his strict standard, AGI has not yet arrived.

• Schmidhuber: LLMs are far from AGI — the physical world is far harder — SchmidhuberAI · 2026-09-07
• Schmidhuber slams Chamath's 'AGI has arrived': no AGI without mastering the physical world — SchmidhuberAI · 2026-09-08
• Schmidhuber: No AGI without self-replicating AI robots mastering the physical world — SchmidhuberAI · 2026-09-08
• Schmidhuber: No True AGI Without Self-Replicating Robot Hardware — SchmidhuberAI · 2026-09-08
• Schmidhuber: no AGI without real-world mastery and self-improving hardware — Mindrust · 2026-09-08
• Schmidhuber: No AGI without mastering the real world — current AI can't match a monkey — SchmidhuberAI · 2026-09-08
• Schmidhuber: AI-controlled general robots could achieve physical self-replication — SchmidhuberAI · 2026-09-08
• Schmidhuber rebuts 'AGI has arrived': no true AGI without self-improving hardware — SchmidhuberAI · 2026-09-09