The mad “gold rush” mentality towards AGI is nerve-racking. I’m reminded of Protogen’s attitude towards the Protomolecule in The Expanse.
I figured we still had 5-ish years to figure it out, but the rapid progress against HLE (Humanity’s Last Exam) makes me nervous.
But sure, let’s just rush headlong towards the precipice, how hard can alignment be really? My anxiety about the future and the importance of getting this right are not eased by people scoffing because “just count the fingers!” When the field is changing so fast, looking at what was going on a few years ago isn’t helpful.
Well, past my pay grade.
As a fellow pedant, I have to point out that even a simple tic-tac-toe algorithm is “AI”.
The term AI was coined at the Dartmouth College Summer Workshop in 1956. Early AI focused on developing expert systems and things like heuristics.
Most people conflate AI, the technical term for computerized decision making in general with the SciFi concept of super intelligent computers, and there has been a revolution since about 2010, in that computationally intensive neutral networks that were theoretical became more conceivable and practical. But LLMs are just a single family of AI techniques.
This, even bad 90’s game computer AI is just as valid to call AI as the latest OpenAI model. It’s just more primitive. Orders of magnitude more primitive, and no neural networks or LLM.