## Due today ```tasks due today not done ``` - [x] ✅ [[2024-09-19]] - [x] deal with is email Dear Coordinators, ✅ [[2024-09-19]] We kindly ask you to forward this communication regarding the teaching sheets to the members of your CCS. The start date of the second semester lessons must be entered on the teaching sheets by February 5th; after this date the forms will be closed for compilation and will then be reopened on 19 February (starting date of the second semester). Teachers will have to enter the start date by referring to the draft timetable drawn up in December. Any changes will be agreed upon and if they occur after the window closes, the office itself will change the date. Thank you for your cooperation and best regards. - [x] Find a place for this academic note creator ✅ [[2024-09-19]] https://chat.openai.com/g/g-fOBxpzt2j-obsidian-integrator-pro/c/fb65e2b3-26d7-4ee9-bba6-68dbeb3ed3eb Key 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 # [[ChatGPT and Skeletal Content]] From https://arc.net/l/quote/visqasyt 1. **Basic Linguistic Patterns and Structures**: - The skeletal content of ChatGPT would comprise the fundamental linguistic patterns and structures it 'learns' during training. This includes basic syntactic rules, common phrase structures, and simple grammatical relationships. These are akin to the patterns of light and dark in photographs or basic sound patterns in audio recordings, representing the foundational layer of language understanding. 2. **Statistical Relationships Between Words**: - A critical aspect of ChatGPT’s training involves understanding the statistical relationships between words, phrases, and sentences. The skeletal content here would be these basic associations – which words are likely to follow others, common word pairings, typical sentence structures, etc., that the model records through its exposure to vast amounts of text data. 3. **Encoded Semantic Principles**: - Although ChatGPT doesn't 'understand' semantics in the human sense, it develops an encoded form of semantic knowledge – basic patterns that indicate how words and phrases are typically used in relation to each other to convey meaning. This encoded semantic knowledge forms part of the skeletal content, underpinning the model's ability to generate coherent and contextually appropriate responses. 4. **Elementary Representations of Context**: - The model also encodes elementary representations of context, understanding, for example, how the meaning of a word can change depending on its placement in a sentence or its relationship to surrounding words. This aspect of skeletal content is crucial for the model to generate responses that are not only grammatically correct but also contextually relevant. 5. **Basic Conceptual Categories and Relationships**: - During training, ChatGPT encounters and encodes basic conceptual categories (like objects, actions, qualities) and their typical relationships. This allows the model to construct rudimentary conceptual mappings, essential for generating text that is logically coherent within the bounds of common knowledge and shared human experience. It's important to note that this analogy stretches [[the concept]] of recording as Kulvicki describes it. In traditional recording, [[the process]] is mechanical and witless, capturing direct sensory input. In contrast, ChatGPT’s 'recording' process is an abstract, algorithm-driven capturing of linguistic patterns and structures. The model doesn’t capture these patterns through direct sensory experience but through computational processes that analyze and encode patterns found in the input data. The 'skeletal content' of ChatGPT, therefore, is a metaphorical construct, representing the most fundamental layers of linguistic knowledge that the model builds upon to generate responses. # [[Photography and Encoding vs Recording]] 1. **Photograph as a Recording**: - **Direct Capture**: A photograph is considered a recording because it involves the direct capture of visual information from the environment. The camera's lens and sensor (or film, in analog photography) register patterns of light and shadow, color, and brightness that correspond to the scene in front of the camera. - **Witless Process**: This process is witless in Kulvicki's sense; the camera doesn't 'understand' or interpret what it captures. It mechanically transduces the visual information into a different medium (digital file or photographic film), preserving the patterns of light and dark, color, etc., as they were received. 2. **Photograph as an Encoding**: - **Transformation of Information**: If we consider encoding in a broader sense as [[the transformation]] and representation of information in a different format or medium, then a photograph can also be seen as an encoding. The visual scene is transformed into a two-dimensional image, where three-dimensional spatial relationships, light, and color are represented on a flat surface. - **Selective Representation**: Moreover, encoding often involves selective representation or abstraction. A photograph abstracts a three-dimensional scene into two dimensions and may involve choices about framing, focus, exposure, etc., which determine how the scene's information is represented in the final image. - **Encoding in the Medium**: In analog photography, the chemical properties of the film encode the light information, and in digital photography, the sensor encodes it into a grid of pixels, each pixel representing color and brightness information. This process transforms the continuous visual scene into a discrete representation, whether chemical or digital. 3. **Analyzing the Duality**: - **Recording and Encoding as Processes**: Philosophically, we can argue that recording and encoding are not mutually exclusive but represent different aspects or stages of [[the process]] of creating a photograph. The recording aspect emphasises the direct, mechanical capture of visual information, while the encoding aspect emphasises [[the transformation]] and representation of that information within a specific medium. - **Intent and Interpretation**: While the camera itself is a witless device, the photographer's intent and the subsequent interpretation by viewers add layers of meaning to the photograph. This introduces another level of complexity: the photograph as a physical record (a recording) and as a piece of visual communication (an encoding of meaning, intended or otherwise). ## [[Taking Photos vs Training ChatGPT]] - - **Recording**: In both cases, this involves capturing information from a given 'scene' (literal in photography, data-based in ChatGPT). However, [[the nature]] of what is captured (light patterns vs. linguistic patterns) and [[the process]] of capture (direct vs. abstract) differ significantly. - **Encoding**: Both processes involve transforming the recorded information into a different format. In photography, it’s the conversion of a visual scene into a chemical or pixel-based image. In ChatGPT, it’s the conversion of text data into patterns within a neural network. - **Representing**: The end product of both processes represents the original source in a transformed manner. A photograph represents a visual scene, and ChatGPT's responses represent the linguistic patterns learned from its [[training data]]. https://www.theverge.com/24027198/adobe-dana-rao-ai-copyright-fair-use-figma-acquisition-deal-decoder-interview ### Notes on [[Moral Realism, Aesthetic Realism, and the Asymmetry Claim]]* Louise Hanson ![[Hanson Ethics Aesthetics 1.pdf]]