IDK if his has been shared in these threads but I came across it recently when someone claimed AGI is coming maybe as soon as next year or 2040 at the latest. Suffice it to say I'm skeptical.
https://voltureluxe.com/ai/how-close...ctually-shows/
The article is almost a year old but I doubt AGI really is just around the corner.
The Intelligence Stack:
We’re pouring billions into making Layer 1 and 2 bigger and faster. But AGI won’t be born from a larger dataset. It will emerge from a breakthrough in layers 3 through 6. Until we shift our focus from mimicry to mind, we’re just building a more sophisticated parrot.
- Computation/Memory (Where AI excels): Raw processing power. The ability to store and recall vast amounts of information. This is what LLMs are built on.
- Pattern Matching/Recognition (Where AI is strong): Identifying patterns in data, from language syntax to visual objects.
- Abstract Reasoning (Where AI struggles): Applying knowledge to novel situations. True problem-solving. This is the focus of benchmarks like ARC-AGI, where the best models still only score around 55%, while humans are near-perfect.
- Causal & Counterfactual Reasoning (Where AI fails): Understanding cause and effect. Asking “what if?” This is the foundation of scientific thought and strategic planning.
- Embodied Cognition (The Missing Link): Learning and reasoning through physical interaction with the world. An AI that has never felt the heat of a fire cannot truly understand the concept of danger.
- Social/Emotional Intelligence (The Final Frontier): Navigating complex social dynamics, understanding unspoken rules, empathy, and building trust.
We’re pouring billions into making Layer 1 and 2 bigger and faster. But AGI won’t be born from a larger dataset. It will emerge from a breakthrough in layers 3 through 6. Until we shift our focus from mimicry to mind, we’re just building a more sophisticated parrot.
The article is almost a year old but I doubt AGI really is just around the corner.

Comment