Member-only story
Benchmarking Human Intelligence versus AI
DeepMind’s Benchmarking of AGI looks like a breakthrough
Tony Czarnecki, Sustensis
London 2/12/2023
For the last two weeks those who follow the events in the area of AI have been preoccupied with the OpenAI saga. We still do not know what was the exact reason for firing and then re-instating Sam Altman as OpenAI’s CEO. However, gradually it is becoming clear that the board’s rebellion was not just about money but about another breakthrough at OpenAI. It is an apparent discovery of a pathway towards Artificial General Intelligence (AGI). That supposedly happened by combining two OpenAI’s teams — the Code Team and the Maths Team. By using a new algorithm called Q* (Q-Star) they may have triggered new capabilities of GPT-4 soon to become GTP-5, alluded by Altman to be a near-AGI.
Almost at the same time something more substantial happened, this time fully revealed, at Google’s Deep Mind, which published an article titled: “Levels of AGI: Operationalizing Progress on the Path to AGI” [1]. Surprisingly, in this breakthrough article, the authors did not define what AGI is. So, before I continue referring to key concepts in this important article, let me fill in the gap, since this term has not been universally agreed yet, by proposing the following definition of AGI:
