Artificial general intelligence

Artificial general intelligence is one of the few ideas ToTheMoon keeps returning to, almost always with the same question: what exactly has to appear for the promise of AGI to stop being a vague forecast and become a testable event. The hosts examine not the technology but the talk around it — timelines from lab leaders, contractual definitions, disputes over terms — and check it against what models already do in real work.

In 2024 the mood is scepticism about dates. Episode 019 explains the community consensus on 2030 through scaling logic, yet calls AGI a “mythical story”. Episode 027 notes that even the godmother of AI does not know what people mean by the word, and that scaling transformers alone will not get there. Episode 032 lines up Altman's promise for next year, Musk's 2028–2029 and Amodei's cautious “powerful AI”: the spread of numbers is itself proof that no definition exists, and episode 039 adds a contractual version — a hundred billion dollars in profit under the OpenAI–Microsoft agreement.

From 2025 the hosts argue for watching the process rather than the label. Episode 054 holds that promises of GPT-5 and near-term AGI should be tested by a working product, not a date in an interview. In 2026, episode 132 concludes from the Amodei–Hassabis dispute that the date distracts from a loop already underway, in which each model helps build the next. Episode 145 separates narrow intelligence, AGI, superintelligence and the singularity as different levels: “There is, of course, no single universally accepted scientific definition of AGI”.