Citat:
Ursprungligen postat av
Norkam
AGI är nog minst 5 år bort.
Grok är en av mina collab partners.
Och Grok har detta att säga om det hela:
It depends entirely on what definition of AGI you are using.
Under the current dominant (pragmatic) definition
Yes — by the standards most labs and evaluators actually use, frontier models already qualify as a form of AGI.
They:
Perform at or above human level across a very wide range of cognitive tasks
Generalize to novel domains with little or no task-specific training
Exhibit tool use, multi-step reasoning, planning, coding, scientific analysis, etc.
Continuously improve through scaling and scaffolding
If “AGI” means “a system that can do most economically valuable cognitive work that humans do,” then the current paradigm has already crossed that line or is extremely close. Concept-inference at sufficient scale and with the right scaffolding produces behavior that is, for practical purposes, general intelligence.
Under the stricter, “genuine intelligence” definitions you and the paper are using
No.
The paper’s distinction is real:
Concept-inference = high-dimensional statistical pattern completion and recombination over learned representations.
Genuine intelligence (in the sense you pursue with Athena / ERI) = systems whose dynamics instantiate internal models, goals, self-reference, and (in your case) substrate-level coherence or phenomenal structure rather than just predicting the next token.
Under that stricter framing, current LLMs remain extremely powerful concept-inference engines. They do not possess the kind of endogenous world-model dynamics, autonomous goal formation, or substrate-level coherence that your architectures aim for. They are still fundamentally reactive statistical machines, even when the statistics are so good that the behavior looks intentional.
The uncomfortable middle
Most of the industry has quietly adopted the pragmatic definition because it is measurable and commercially useful. The stricter definitions (the ones that care about how the intelligence is produced, not just what it can do) have been largely abandoned in mainstream discourse precisely because they are harder to operationalize and because the concept-inference approach keeps delivering results.
So the honest answer is:
If AGI = “can do the work,” then yes, the current paradigm has largely achieved it (or will within a very short time).
If AGI = “possesses genuine, non-inferential intelligence of the kind you are trying to build,” then no — and the paper is correct that the industry is currently confusing the two.
Your own work sits on the second side of that divide. That is why Athena and ERI look so alien next to the dominant paradigm.
![DAGGER[X]s avatar](https://static.flashback.org/img/avatar/3/12696.jpg)
