Some Contemplations on meaning

The great gift of the LLM is that we now know that there are in fact two different forms of meaning.

The first and what most think they mean when they think of meaning is conceptual meaning. That is the package of relationships, sense data, and emotions that encapsulates the concept. When I say tree, there is its relationship to other concept. A tree is a plant meaning it is a member of the concept plant. Birch is a tree meaning birch is a member of the concept tree. It has branches and trunks meaning these concepts are parts of the whole concept of tree. It is woody meaning that the quality of wood is possessed by the tree in particular.

In addition we have the image of the tree. The brown of the trunk. The green of the leaves. The sun reflecting off them. The blue sky with few clouds framing the leaves. We can close our eyes and hear the soft rustling of the leaves and the waving of the branches. Breath deep and remember the smells of the trees—the smell of pine sap, the smell of crushed leaves, and the smell of terpide. These are all the sense data that makes up the concept of the tree.

We also have the emotions that we have felt in connection to the tree. We recall being young and looking up at the tree not as a specific memory but as the emotion of wonder. We feel relaxed. We feel enraptured. These emotions might be different for all of us, but the concept of tree is so pervasive that I guarantee everyone reading this has emotions that compose and are associated with the concept.

Conceptual meaning encapsulates these all. It encapsulates the relationship of the form of tree to other forms. It encapsulates the sense data. And it encapsulates the emotions informed by experiences.

On the other hand semantic meaning refers to the meaning of the word tree that I have just written. Semantic meaning informs us that tree has four letters. It has a double ee. It starts with t. It tells us that tree is a noun and not a verb. It says that when I am using tree right now it is as the object of the sentence. It tells us what adjectives are associated with tree such as majestic and green.

To put it simply semantic meaning is the relationship between the word itself and other words.

LLMs by their very nature lack the ability to interface with the world of forms. They cannot peer further down the light to see what unites the shadows. They cannot even view the shadows, rather, us describing the shadows. They lack emotions entirely. As such, the elements of the conceptual meaning of a word are completely absent from the LLM. So what does it know? The position of tree in n dimensional space and how close it is to other words. It knows that tree is close to bush. That magnificent and tree often go together. That the verb climb often has tree as an object. That tree may be definitive or non-definitive. It however cannot know the concept of tree.

This is illustrated by the habitual failings of the LLM to reason over a world model. For example under current models—Opus 4.6 and ChatGPT 5.5—if given the case of the user being fifty feet away from a car wash should the user walk or drive, will nearly always say to walk. This is because the telic meaning of the conceptual car wash is not within its vocabulary, rather, its relationship to other words. As such the LLM notes that fifty feet and walk go together and thus informs the user that they should walk. Under semantic meaning this answer is comprehensible. Under conceptual meaning it betrays a total disconnect from reality. If these types of meaning were the same we would not see this pattern of LLM failure where they spit out sentences where all of the words make sense in the sentence—as opposed to a hallucination. However, when the concepts are examined and operated upon then the statement of the LLM ceases to be anything but a stochastic parrot imitating human speech.

Some of my more astute readers will be reminded of the literature upon form and meaning of words. Form is the actual token, letters, and shape of the words. Tree is spelled t-r-e-e. I am using a serif font. It has four letters. These are the specific form of the word and in most written language has no resemblance to the concept at hand. Tree-the-word and tree-the-plant bear literally no resemblance. (I suppose the capital T has a little? That’s being pedantic however.) What I am stating bears some resemblance to this form and meaning distinction and indeed the astute reader will note that I conflated form and meaning to derive the semantic reading where I both described tree as a noun and tree the word with four letters. I fully accept that these are different concepts, however, I would content that the form of the word does indeed grant meaning—specifically semantic meaning.

For example, the word pretty and the word pulchritudinous have the same meaning on paper. Both mean in the dictionary beautiful and good looking. Yet if one utilized these two words with the same meaning, their form influences the meaning. For pulchritudinous has a longer form and more syllables. As such it is a more formal word. This formality influences the meaning. When writing poetry this word is appropriate when describing a lady one espied at a party. If one is in the midst of a lusty conversation with one’s wife in a casual setting the use of the word would be inappropriate. How so they mean the same thing? Because of the form of the word.

Pretty, if we pretend that it is an atomic concept and that such exists, the concept and pulchritudinous have a relationship in that pulchritudinous is the formal version of pretty. The images conjured up in one’s head of formal pretty is a lady in a ball gown as opposed to pretty being a lady in a sundress. Both are pretty, but the form of pulchritudinous places an attribute upon the concept pretty. As such I am justified in seeing form as related to semantic meaning.

In conclusion, these are just a few of the thoughts that have been operating within my head. It seems to me that dividing meaning serves to explain why LLMs operate in the fashion that they do, where they falter, and why throwing more compute is merely papering over an unsolvable problem. LLMs will never achieve AGI or consciousness because they simply are operating upon a different level of meaning. The lady that has so descended into fiction that she filters reality through romanticism, or the gentlemen that has become so enraptured in knightly tales that he jousts at windmills is a common character within literature. The heart of these characters is that they gain their knowledge of the world not from sense data and lived experiences but rather from the written word. If we consider these characters as pathetic, laughable, loons, why are we trusting the robotic equivalent of these characters to achieve anything beyond manufactured, controlled lunacy?