IAIRC

We often try to understand animal intelligence, artificial intelligence, and other forms of intelligence by comparing them to ourselves. We ask whether they think like humans, reason like humans, learn like humans, or can perform tasks at a human level. Instead we need to ask ourself:

An intelligence does not necessarily need to think like us, learn like us, or experience the world like us to have significant consequences for humans.

Understanding intelligence therefore requires more than measuring its similarity to human intelligence. We should study what kind of intelligence a system possesses, what it is capable of within its environment, and what effects those capabilities have on human life and as that intelligence becomes more capable, we should also study how the scale and nature of its effects on humans change.

Transformers have demonstrated remarkable capabilities, but their success does not necessarily mean that scaling them will lead to AGI.

IAIRC explores the possibility that achieving AGI may require new architectures, mechanisms, or approaches to intelligence beyond today's transformer-based systems.