What do I need to do right now? I need to think of an idea for the chatgpt metaphysics paper. [[What can]] it be about? Start with TC because he is the most interesting. [[Ted Chiang]] thinks of chat GPT as a "Blurry JPEG of the Web". Open AI have scanned a vast amount of data (let's assume most of the internet) and instead of storing it have made a lossy compressed copy: Think of ChatGPT as a blurry _JPEG_ of all the text on the Web. It retains much of the information on the Web, in the same way that a _JPEG_ retains much of the information of a higher-resolution image, but, if you’re looking for an exact sequence of bits, you won’t find it; all you will ever get is an approximation. But, because the approximation is presented in the form of grammatical text, which ChatGPT excels at creating, it’s usually acceptable. You’re still looking at a blurry _JPEG_, but the blurriness occurs in a way that doesn’t make the picture as a whole look less sharp. TC describes this story as "an analogy" –how close is this analogy to how OpenAI actually trained ChatGPT? **ChatGPT itself says...** Does the *way* in which something functions have an effect on what it is? If we define artifacts purely by their function then no(?) Let's assume it is. [[What sort]] of thing does that make chatGPT? Could we take it literally? An image of the internet? How would that work? The most obvious difference would be how we access the information that *is* in the image. *Attending to different parts of the picture vs. typing in commands* However, a difference in interface does not necessarily indicate a difference in *type of thing*. Think of different types of telescope/apparatus being used to look at a heavenly body/different types of microscope being used to look at something small. Maybe its better to think of chatgpt as an image of information rather than an image of text? WHat would this amount to? As we dial down the repleteness, for example, we move from pictures to certain kinds of graphs, say, of temperature [[over time]], or temperature across a surface, and so on. If we give up on syntactic density, or on a high degree of sensitivity, we wind up closer to linguistic representations and some diagrams, which are easily distinguished from one another, instead of being indefinitely sensitive to changes in their features. # The Enrico Meeting Possible Title: The place of LLMs in the realm of digital artifacts. - A neutral characterisation of artifacts: structure is the matter, function is the form - What is the function of LLMs? - structure involves a series of capacities, function is the use that is selected (still remaining neutral) ## The Computer - computers are concrete artifacts. - What is [[the structure]] of the computer? CPU + Memory + Input + Output - Structure of Layers? - What is the function of [[the structure]] is to execute programs (programs are digital artifacts) - Two main kinds of digital artifacts programs and data. - Usually a digital aritfact is a file. ## Programs - structure of programs = data + instructions - function of a program = inputs leading to outputs - programs can be compared to corporeal artifacts - Another distinction is between the programs that have specified outputs to inputs, and those like LLMs which have less that specified outputs ## Networks - [[The structure]] of a computer network: Wires, TCP - The function of a network, a bit like computers in that it is the running of programs - Client/Serve Programs - Philosophy of computer science –Fioridi, has he written this sort of stuff? It is surprising that no one has ## What is the function of a browser? ## chatGPT as similar to the imagination? ## scrap from gpt Ok good. Now I am going to ask you quite a challenging question so be careful to think in steps when you answer, remember what a good philosopher you are. My question is as follows: [[Ted Chiang]] says: "Think of ChatGPT as a blurry _JPEG_ of all the text on the Web. It retains much of the information on the Web, in the same way that a _JPEG_ retains much of the information of a higher-resolution image, but, if you’re looking for an exact sequence of bits, you won’t find it; all you will ever get is an approximation. But, because the approximation is presented in the form of grammatical text, which ChatGPT excels at creating, it’s usually acceptable. You’re still looking at a blurry _JPEG_, but the blurriness occurs in a way that doesn’t make the picture as a whole look less sharp." Kulvicki is happy to say that there can be non-visual pictures, so do you think we can flesh out [[Ted Chiang]]'s idea in terms of Kulvicki's four features? Give me plenty of detail, anddon't be afraid to talk about the negative as well as the positve.