Note! This is an extremely biased and subjective personal record, not of the technical or feature-level aspects of LLMs, but of a shift in the range of human cognition over time. Personally, I do not think I had ever left a review of a technology while doing Coding Everybody-style exercises. But while working through the GPT-4o API practice, especially the problem of finding the area of a shape drawn with lines, I was hit by no small shock. So this time I am leaving behind some personal reflections from that process, reflections that may feel embarrassing when I read them later.
A review of OpenAI GPT-4o API practice
1. First impression, universal access to multimodality
? 2. The background of that feeling, I think therefore I am (feat. hallucination)
Book | Discourse on the Method
René Descartes
There are two ways by which we arrive at knowledge of things: experience and deduction. Experience can often fall into error, whereas deduction, that is, pure reasoning that draws one thing from another, can fail only if we are not careful, since it is performed by the understanding, or by reason itself.
What matters here is that even if sense, especially sight, appears to present external objects exactly as they are, the understanding is never deceived by experience itself, unless it judges that external things are truly as they appear. We fall into error precisely when we make that judgment.
A portion of something I wrote in the past
Part of what I wrote last time in a post about visual perception
Vision arrives twisted, warped, overlapped, and inverted. I hope people understand clearly that visual perception is not something that happens automatically but something done actively by investing cost. (Choi Nak-eon)
And below is part of something I organized last year, in 2023, while running a book study on UX Psychology at From Designer.
It is an excerpt from notes I made while looking for materials on vision-centered attention and depth perception.
Book | What Is an Image?
Lee Sol
Recognition through experience falls into error because we make hasty judgments about the experience that is given to us. The problem is believing, without doubt, that the given experience is simply fact. For example, the fact that a stick in a glass of water looks bent is not itself the problem. The error arises when we accept that visual experience as it is and judge that "the stick is bent." Error lies not in subjective experience itself, but in the mistaken pairing of subjective experience with objective reality.
We cannot guarantee that the image given to the subject through experience properly reflects reality. The moon in the night sky appears smaller than a tall building, and scenery seen from a fast-moving car seems to move backward. Since our experience often commits errors in this way, experience can in principle always commit errors.
The errors that occur while an LLM AI such as ChatGPT processes information are called hallucinations. But when I look at the content above, I start to think that perhaps hallucination is something only humans can truly do, in other words bias or cognitive dissonance. LLMs can relatively easily adjust past errors through fine-tuning, RAG, or new prompts. Humans, by contrast, face great difficulty in adding new information or revising past judgments on the basis of new information.
Book | Meditations
René Descartes
I am now staring at this paper with my eyes wide open.
This head that I move here and there is not asleep. I deliberately stretch out my hand and I feel it as well. When I was asleep, such things were not as clear as this. And yet how many times have I not been deceived by similar thoughts in dreams? If I think carefully about this, I am startled to realize that there is no certain sign by which being awake can be distinguished from dreaming. (Meditations, 36)
I am the cogito.
Now let me close my eyes, stop my ears, push away all the senses, and erase from my thought all images of material things. If that is too difficult, let me at least regard such images as empty and false and ignore them. Let me speak only with myself, examine my inner self more deeply, and make myself gradually more known and more familiar to myself. I am a thinking thing. (Meditations, 56)
The existence of myself as a thinking being becomes the foundation of all knowledge. Paradoxically, this first knowledge is discovered on the very fact that I am being deceived. Of course, an all-powerful evil demon may deceive me through every kind of illusion. But for that demon to deceive me, there must at least be a me who is being deceived. The first form of myself that Descartes discovers is thus this self that is being deceived, this self that has false representations. "I am, I exist." (Meditations, 44)
Here again, we discover that the question, "What if this is not real but a dream?" already presupposes a prejudice about images. Descartes separates the image given to consciousness from the reality outside consciousness.
The image within consciousness clearly imitates reality, but there is no clear standard by which to determine whether it properly corresponds to reality. The anxiety that what is given to my consciousness may be a false representation, the anxiety expressed in the form "Is this perhaps a dream?" moves with the hypothesis of the evil demon toward an extreme skepticism about all knowledge.
Interestingly, Descartes chooses to accept the problematic situation itself. He accepts the possibility that all the images given to him may be false, and admits that he can no longer distinguish truth from falsehood. "Very well then, let us suppose that we are now dreaming." (Meditations, 36)
I who doubt whether the images given to me are true or false exist. Regardless of whether those images are true or false, the existence of the self that is thinking about these true or false representations is certain. It is precisely in this sense that "I am a thinking thing. (Ego sum res cogitans)" (Meditations, 56) becomes the first certain knowledge Descartes discovers while turning away from illusions and appearances and gazing inward. By returning to himself, Descartes reaches the truth he long longed for. Truth dwells not somewhere else, but within his own interior.
In Discourse on the Method, Descartes writes, "I think, therefore I am" (Je pense, donc je suis), and in Principles of Philosophy he presents this as "Ego cogito, ergo sum."
