ChatGPT: AGI by 2024, the hard part is now done

LLM-based prototype AGI may come well before the 2040 median expert prediction. IMAGE CREDIT | Adapted from waitbutwhy.com

ChatGPT is not yet AGI, so-called, artificial general intelligence. And it may have some fundamental limitations. But here I argue we might get early GPT-based AGI in 2023 and decent ones by 2024. Let’s talk about how that might happen.

Here’s a few quick assumptions I make:

1. My definition of AGI is a human-like independently operating entity with apparent cognition, common sense, somewhat human-level knowledge PLUS persistent short and long-term memory, capability to follow multi-step instructions and has apparent goals, interests and pseudo-emotions
2. A first useful AGI need only display smart two-way text interaction. Other senses are here considered useful but not critical: interaction via images, sounds, smells, tastes, touch, speech or limbs.
3. AGI occurs across a spectrum: from ‘error prone’, not-too smart or ‘savant’-like (sub-human reliability or intelligence of limited scope) all the way to human-level intelligence and then on to super-intelligence.

Such a nascent AGI could soon aid and manage our projects and even businesses, at least in shadow mode or for short periods.

Additionally, we will assume that present LLMs (large language models) like GPT-3 / ChatGPT do at least deliver apparent human-level understanding reasonably…