What a bullshit argument. One of the arguments for self driving cars is precisely that they are not doing the same thing humans do. And why should they? It’s ludicrous for a company to train them on “social norms” rather than the actual laws of the road. At least when it comes to black and white issues as what is described in the article.
It’s like my brain is not dissimilar from the toddler I have at home. If I don’t take charge, come up with t activities, give it a routine, feed it properly, make sure it is hydrated and well slept it turns into a bored, bossy, moody asshole. Of course they, both toddler and brain, still do sometimes but if I do all the things above I’m usually prepared. Sometimes they also are super good at doing all the things by themselves.
In some ways, while extremely exhausting, having a toddler is actually great for my brain because otherwise I would not eat well, make plans, follow routines, go out, or have fun. My brain is still unhappy of course because it wants to do nothing and eat “chips with sugar” (to quote this toddler being I have at home).
I’m not sure if you’re disagreeing with the essay or not? But in any case what you’re describing is in the same vein, that is simply repeating a word without knowing what it actually means in context is exactly what LLMs do. They can get pretty good at getting it right most of the times but without actually being able to learn the concept and context of ‘table’ they will never be able to use it correctly 100% of the time. Or even more importantly for AGI apply reason and critical thinking. Much like a child repeating a word without much clue what it actually means.
Just for fun, this is what Gemini has to say:
Here’s a breakdown of why this “parrot-like” behavior hinders true AI:
- Lack of Conceptual Grounding: LLMs excel at statistical associations. They learn to predict the next word in a sequence based on massive amounts of text data. However, this doesn’t translate to understanding the underlying meaning or implications of those words.
- Limited Generalization: A child learning “table” can apply that knowledge to various scenarios – a dining table, a coffee table, a work table. LLMs struggle to generalize, often getting tripped up by subtle shifts in context or nuanced language.
- Inability for Reasoning and Critical Thinking: True intelligence involves not just recognizing patterns but also applying logic, identifying cause and effect, and drawing inferences. LLMs, while impressive in their own right, fall short in these areas.
Interesting you focus on language. Because that’s exactly what LLMs cannot understand. There’s no LLM that actually has a concept of the meaning of words. Here’s an excellent essay illustrating my point.
The fundamental problem is that deep learning ignores a core finding of cognitive science: sophisticated use of language relies upon world models and abstract representations. Systems like LLMs, which train on text-only data and use statistical learning to predict words, cannot understand language for two key reasons: first, even with vast scale, their training and data do not have the required information; and second, LLMs lack the world-modeling and symbolic reasoning systems that underpin the most important aspects of human language.
The data that LLMs rely upon has a fundamental problem: it is entirely linguistic. All LMs receive are streams of symbols detached from their referents, and all they can do is find predictive patterns in those streams. But critically, understanding language requires having a grasp of the situation in the external world, representing other agents with their emotions and motivations, and connecting all of these factors to syntactic structures and semantic terms. Since LLMs rely solely on text data that is not grounded in any external or extra-linguistic representation, the models are stuck within the system of language, and thus cannot understand it. This is the symbol grounding problem: with access to just formal symbol system, one cannot figure out what these symbols are connected to outside the system (Harnad, 1990). Syntax alone is not enough to infer semantics. Training on just the form of language can allow LLMs to leverage artifacts in the data, but “cannot in principle lead to the learning of meaning” (Bender & Koller, 2020). Without any extralinguistic grounding, LLMs will inevitably misuse words, fail to pick up communicative intents, and misunderstand language.
This was my biggest takeaway here. Wtf?! “I personally set the price and thought we would make some money”?! Either he is trying to sound cool by being casual or he is a fucking idiot. Or probably both.