Descriptions:
TheAIGRID breaks down a viral statement from Demis Hassabis, CEO of Google DeepMind, who declared that today’s AI systems are “nowhere near AGI” — a notable position given the recent wave of claims that artificial general intelligence has effectively arrived. Hassabis’s argument is that AGI requires the full range of human cognitive abilities: invention, planning, reliability, and creativity across all domains, not just impressive performance on narrow tasks.
The video frames this against OpenAI’s announcement that an internal model disproved a conjecture connected to Paul Erdős’s 1946 planar unit distance problem, with the proof verified by external mathematicians. Hassabis is presented as arguing that while this is impressive, it’s evidence of progress — not proof that AGI has been reached. Gary Marcus is also cited, pointing to persistent quirks in frontier models (including GPT-5.5’s need for explicit system-prompt instructions about “goblins”) as evidence that current systems remain fundamentally brittle.
The counterarguments get serious treatment too. Marc Andreessen’s position — that models already provide better answers than top experts across medicine, law, coding, and finance, and that this is what general intelligence looks like in practice — is presented as credible. Andrej Karpathy’s concept of “jagged intelligence” provides the central explanatory frame: AI has strengths and weaknesses that don’t correlate the way human cognitive abilities do, making it simultaneously impressive and unreliable in ways that are difficult to predict in advance.
📺 Source: TheAIGRID · Published May 28, 2026
🏷️ Format: News Analysis
