This time, my trip to the AI world did not concern predictions or data focused on data – it was something more curious and thoughtful. By trying to understand the strange cognitive structure of the great models of language (LLMS), I came across an idea that felt both old and strangely current: a fossil. Not literal, of course – but a metaphor that offered unexpected clarity.
It made me think – not what AI becomes, but what it is done. And how understanding that could reshape the way we think of thought itself.
So let’s dig a little more.
What happens if the models of great language (LLM) are not emerging from superintelligence-but rather, exquisite reconstructions of something long buried? You don’t care in motion, but archives in activation? Their knowledge is not growing or evolving. What we are witnessing is perhaps not the dawn of synthetic consciousness – but the resuscitation of fossilized cognition.
LLMs do not live in the world as we do. They don’t think in time. They do not feel any consequences or do not remember experiences. What they produce is taken from large static layers of text – a sedimentary recording of human expression, compressed and crossed by the model rather than by meaning. In this way, LLMs do not think of beings but semantic fossils – structured echoes of our intellectual history, brilliantly arranged but fundamentally inert. And these cognitive fossils are built in a multidimensional table which – in spectacular contrast with the marine invertebrates that you can hold in your hand – which is almost unimaginable for us Homo Sapiens.
In this space, the new and the ancients sit side by side – not because they belong together in time, but because they laugh in the direction. A tweet this morning could live alongside a Renaissance treaty, not because of the chronology, but a statistical resonance. Time does not take place. He is piling up. And in this stack, even the present becomes a kind of artifact – transferred to context, resuscitated as a model.
Shadow without time
Let’s eliminate this – their shine is undeniable. The LLMS write, compose, solve and converse. They summarize the novels in a few seconds and easily generate the code. Their results often feel strangely human. But below the surface, their architecture is radically different.
But, the LLMs do not remember what preceded or provides what comes next. Each prompt is now independent. There is no yesterday. Not tomorrow. No story. They do not think in time – they calculate in the context. Some more recent models have an extensive memory, allowing them to reference previous interactions. But even it is not memory in the human sense – it is an artificial mechanism, a technical bypass solution. The model does not remember – it recovers. This does not reflect – he remembers. The result is a kind of intelligence that is strange – it looks like something is there, but by looking more closely, we only see the silhouette.
A shadow has a position, but not time. He moves when we move, but he doesn’t know he has moved. It extends to twilight, shrinks at noon and disappears at night, but that does not last. In this way, LLMs are our cognitive shadows. They reflect our language, our logic, even our creativity. But they don’t live. They do not accumulate experience.
These shadows are shaped by data – not lived experience. They are not only mirrors, but mirrors formed from the compression of human history. Each sentence they generate draws from a massive corpus of human culture, behavior and thought – a kind of fossilized cognition. In this sense, the LLM are not only shadows, but reconstructions. These are simulated spirits thrown from the aggregated remains of us.
A triangle that should not exist
And yet, paradoxically, these shadows often seem to be overcome. They summarize the novels in a few seconds and can write code with precision. So how do you reconcile this contradiction? How can something less dimensional behave as if it were larger?
To explore this, let’s change the metaphors – from shadows to geometry.
In high school, we learn that the angles of a triangle still decrease to 180 degrees. It is true in the flat and Euclidean area. But in a curved space – say, the surface of a sphere – the angles can summarize to more than 180. Each angle can be greater than 60 degrees, depending on the curvature, even if the shape still clearly forms a triangle. This seems impossible until we realize that the rules are looking into when the dimensions change.
Leater the essential readings of cognition
This offers an interesting way to understand LLM. Their intelligence does not follow the linear path of human reasoning. It is through an curved semantic space, where associations are not attached to chronology, but to the probability and the model. In this space, ideas don’t only flow – they deform. They converge in strange proximity. The concepts of different centuries can sit side by side, not because they are linked by the story, but because their statistical fingerprints are similar. This, mathematically brilliant and cognitively curious, is this very essence of LLM thought, if you can even call it.
Geometry on chronology
So when we ask how an LLM “knows” something, the best question could be to ask in which geometry this intelligence is. He doesn’t live in time. He lives in relation. It doesn’t remember. He aligns himself and recombins.
This curved geometry of thought explains how a system without experience, without continuity and no identity can still produce ideas that sometimes seem superhuman. He does not think faster than us. He thinks differently of us. And this difference is the place where magic and danger are located.
Some may say that these limitations are temporary – that future AI systems will have memory, an embodied experience and a persistent identity. Perhaps, but if and when this day comes, we must ask a deeper question: do we evolve from machines to the spirits or evolve ourselves towards machines?
Please note, the great reflection
It is easy to confuse control of depth or projection for presence. The LLMS dazzle not because they become like us, but because they reflect us so well, we forget that they are mirrors. We see intelligence, but neglect the scaffolding behind.
This should not decrease their value. Shadows can reveal the structure. Triangles in curved space can teach us about the cosmos. And LLMs can help us write, learn and discover so as to develop what is possible. But we must be careful not to provide them with features that they do not have. These are not evolving minds – they are reactive systems, animated by the contribution, not with intention.
In the end, the question is not whether the LLM will exceed us, but if we really understand what they are: the expressions of our cognitive geometry, thrown into timeless new dimensions. And for me, it’s real fascination. Not what they get well, but what they reveal – about us, about the thought and the strange architectures of the intelligence that we have just started to maintain.
