AI industry leaders continue to offer sharply different timelines and definitions for artificial general intelligence, while newer terms such as “recursive self-improvement” are creating another difficult-to-measure target. A running tally published by PC Gamer on September 26, 2026, highlights how predictions from executives and researchers have repeatedly ranged from AGI already existing to its arrival several years away.
AGI Remains a Moving Target
The latest claim cited by PC Gamer came from Nvidia chief Jensen Huang, who said in September 2026 that AGI had arrived with OpenAI’s GPT-6 Astra, reportedly trained on more than 100,000 Nvidia Grace Blackwell NVLink72 systems. The source provides no benchmark or shared definition establishing that the model meets an accepted AGI threshold.
That position differs from an August 2026 statement attributed to OpenAI chief Sam Altman. He reportedly said OpenAI was not there yet, but expected to have an internal system he would call AGI by the end of the year. Earlier forecasts also varied widely: Elon Musk suggested xAI might achieve AGI as soon as 2026, while Google DeepMind’s Demis Hassabis estimated in January 2025 that it remained three to five years away.
Definitions Shape the Deadlines
The disagreement is partly definitional. PC Gamer’s examples include claims based on exceeding human performance, passing a specified collection of tests or providing broad assistance across cognitive tasks. Without a consistent threshold, an executive can describe AGI as present while another treats it as an internal milestone still under development.
The timeline has shifted repeatedly as well. Altman predicted in November 2024 that AGI would come in 2025, while Musk said in April 2024 that it could arrive within one or two years. Nvidia’s Huang separately forecast in March 2024 that a test-based version could be reached within five years.
A New Frontier Term Emerges
PC Gamer identifies “recursive self-improvement” as a term that gained prominence in June 2026. Anthropic described it as a possible stage in which AI systems autonomously design and develop their successors, while explicitly acknowledging that the capability had not been achieved and was not inevitable.
The terminology is already influencing governance discussions. In July 2026, a statement signed by 1,386 employees of frontier AI companies asked the US government to support an international effort for tools that could deliberately slow automated AI development. The proposal reflects concern about rapid progress, but the source does not establish that recursive self-improvement is currently available or imminent.
AI-generated editorial illustration; not a documentary photograph.
Source: A running tally of the most ridiculous things AI bigwigs have said so far
SIGNAL TITANS
Leave a Reply