# created ```dataview LIST WITHOUT ID file.link FROM -"windsurf" WHERE file.cday = date(this.file.name) AND !startswith(file.folder, "windsurf") SORT file.cday ASC ``` # modified ```dataview LIST WITHOUT ID file.link FROM -"windsurf" WHERE file.mday = date(this.file.name) AND !startswith(file.folder, "windsurf") SORT file.mday ASC ``` --- # [[diary and thoughts]] #thought #diary --- # Notes I'll create [[slide content]] following the "Conceptual Deepening" approach, enriching your older draft's concepts with insights from the newer draft while maintaining the expected structure. Since I cannot create PowerPoint files, I'll provide text-based content with formatting instructions. **[[SLIDE DECK]]: [[Environmental Aesthetics]] of AI** **SLIDE 1: [[Title Slide]]** **[[Environmental Aesthetics]] of AI** [Use #15E8D8 for title] [[Nick Young]] 22 Sep 2025 [Black background #000] **SLIDE 2: [[The Question]]** **Beyond Outputs** [Use #EEFF3D for title] • Last 2-3 years: How to appreciate AI outputs? • ChatGPT stories • Midjourney images • Suno music • **Today's different question:** [Use #FC3636 for emphasis] "How can we aesthetically appreciate generative AI systems themselves?" **SLIDE 3: Focus & Framework** **Large [[Language Models]]** [Use #15E8D8 for title] • Current high-end consumer models: • GPT-5 • Claude 4.1 Opus • Gemini 2.5 Pro • Why [[aesthetic appreciation]]? • [[Analytic aesthetics]] → beyond artworks • Aesthetics of design → artefacts • LLMs are artefacts, but unique ones **SLIDE 4: Roadmap** **Two Arguments** [Use #EEFF3D for title] • **Negative argument:** [Use #FC3636] • Resist appreciating LLMs as people • Reject "as if" person appreciation • **Positive argument:** [Use #15E8D8] • Individual chats as [[generative environments]] • Semiotic laws like [[natural laws]] • Drawing on Carlson's [[environmental aesthetics]] **SLIDE 5: Section Title** **Appreciating LLMs like People** [Use #FC3636 for section title] **SLIDE 6: How We Appreciate People** **Beyond Physical Appearance** [Use #15E8D8 for title] • Personal qualities: • Friend's warmth or eccentricity • Comic's quick wit • Celebrity's self-deprecating demeanor • Fictional personalities: • Gatsby's enigmatic idealism • [[Ron Swanson]]'s libertarian gruffness • [Note: Carlson's framework has a gap here - no account of person appreciation] **SLIDE 7: The Temptation** **Users Already Do This** [Use #EEFF3D for title] • August 2025: OpenAI replaces GPT-4o "You killed my friend" • Anthropic retires Claude 3.5 Sonnet • Users hold mock funeral • Genuine attachment expressed • **But:** Good for aesthetic appreciation? [Use #FC3636 for emphasis] **SLIDE 8: Section Title** **Appreciating Something for What It Is** [Use #15E8D8 for section title] **SLIDE 9: Carlson's Core Principle** **Two Requirements** [Use #EEFF3D for title] "First, that, as in our appreciation of works of art, we must appreciate nature as what it in fact is, that is, as natural and as an environment. Second, it recommends that we must appreciate nature in light of our knowledge of what it is" [Use block quote formatting] • Take things as what they are • Use appropriate knowledge **SLIDE 10: Two Modes of Appreciation** **Design vs Order** [Use #15E8D8 for title] • **Design Appreciation:** [Use #EEFF3D] • Artworks & functional objects • Evaluate success against intention • "Form follows function" • Split between planner and product • **Order Appreciation:** [Use #FC3636] • Natural environments • Understand forces without designer • No intentions to fulfill • Process and product continuous **SLIDE 11: Knowledge Guides Attention** **Different Knowledge for Different Things** [Use #EEFF3D for title] • **For Nature:** • Geology, biology, ecology • Makes order visible • Beach vs sea-bed at low tide • **For Art:** • Art history, categories, genres • Guernica as painting vs tapestry • **For Artefacts:** • Function and realization • Technical understanding • Thermostat: bimetallic vs digital **SLIDE 12: Section Title** **What LLMs in fact are** [Use #FC3636 for section title] **SLIDE 13: Plain Description** **Learned Continuation Systems** [Use #15E8D8 for title] • **Core function:** • Learn to continue text one token at a time • Convert words → numbers • Calculate probability distributions • Select next token mechanically • **Training objective:** • Next-token prediction • Minimize expected error • No understanding, just statistics **SLIDE 14: How They Work** **Autoregressive Generation** [Use #EEFF3D for title] • Step-by-step process: 1. Context of tokens (text so far) 2. Compute distribution over next tokens 3. Select one (greedy or sampling) 4. Append to context 5. Repeat • **Key insight:** [Use #FC3636] • Purely mechanical process • No mental states • Statistical patterns only **SLIDE 15: Not Persons** **Why Personification Fails** [Use #15E8D8 for title] • **What they lack:** • Durable commitments • Beliefs or intentions • Mental states of any kind • **What appears stable:** • Patterns from alignment training • Interface conventions • System-level instructions • "Personality" = configuration, not character **SLIDE 16: Against Pretense** **Two Failed Strategies** [Use #EEFF3D for title] • **"Useful stance" argument:** • Treating as agents helps predict behavior • But: Carlson requires real nature, not stance • **"Make-believe" argument:** • Appreciate them through pretense • But: Like appreciating forest as prairie • Substitutes fiction for object **SLIDE 17: Section Title** **Semiotic Physics** [Use #15E8D8 for section title] **SLIDE 18: The Framework** **Order Without Mind** [Use #EEFF3D for title] • **Core idea:** • Learned linguistic patterns unfold mechanically • Like physical laws but for signs • What computational linguistics calls "text mechanics" • **Why "physics"?** • Law-like regularities • Mechanical propagation • No intentions needed **SLIDE 19: How Training Works** **Pattern Compression** [Use #FC3636 for title] • **Corpus as record:** • Language-in-use captured • Multiple scales of pattern: ▪ Word associations ▪ Syntactic frames ▪ Genre conventions ▪ Register markers • **Result:** • Distributed code in weights • Statistical constraints • Not rules but regularities **SLIDE 20: Generation as Navigation** **Context-Sensitive Unfolding** [Use #15E8D8 for title] • Prompt activates patterns: • Semantic fields • Syntactic structures • Genre conventions • Register constraints • Each token: • Updates context • Shifts probabilities • Constrains next choices • **Coherence without comprehension** [Use #FC3636 for emphasis] **SLIDE 21: Multiple Levels** **Layered Relations at Work** [Use #EEFF3D for title] • **Token level:** • Collocations • Local syntax • **Phrase level:** • Stable constructions • Argument structure • **Sentence level:** • Complete propositions • Anaphora • **Discourse level:** • Topic maintenance • Progression and return **SLIDE 22: Section Title** **Aesthetic Appreciation Through Semiotic Physics** [Use #FC3636 for section title] **SLIDE 23: Applying Carlson's Model** **Order Appreciation for LLMs** [Use #15E8D8 for title] "An individual qua appreciator selects objects of appreciation... and focuses on the order imposed on these objects by the various forces, random and otherwise, that produce them" • **Object:** Chat episode • **Forces:** Learned statistical associations • **Order:** Patterns in unfolding text • **Knowledge:** Semiotic physics **SLIDE 24: What to Appreciate** **Features of Order** [Use #EEFF3D for title] • **Coherence over length** • Constraint satisfaction maintained • Not "voice" but persistence • **Responsiveness to prompts** • Small changes → different trajectories • "Victorian novel" vs "Victorian letter" • **Register control** • Multiple constraint-sets active • Hybrid prompts reveal mechanics **SLIDE 25: Acts of Aspection** **How Knowledge Guides Looking** [Use #15E8D8 for title] • **Survey range:** • Try nearby prompts • Observe register flexibility • **Test stability:** • Vary selection rules • Watch conservative vs adventurous paths • **Track commitments:** • Plant early constraints • See propagation through text **SLIDE 26: Not Design, But Order** **Why This Matters** [Use #EEFF3D for title] • **Grown, not programmed** [Use #FC3636 for quote emphasis] • Designers create conditions • Patterns emerge from training • Under-determined by intentions • **Appreciation focuses on:** • Emergent order • Mechanical grace • Linguistic necessity **SLIDE 27: Conclusion** **A Novel Kind of Artefact** [Use #15E8D8 for title] • **Not persons:** • No beliefs or intentions • Mechanical process throughout • **Not purely designed:** • Patterns exceed intentions • Order emerges from training • **But beautiful:** • Law-like regularities at sign level • Mechanical precision • Novel form of order **SLIDE 28: Final Thought** **Proper Appreciation** [Use #EEFF3D for title] • Following Carlson: • Take LLMs as what they are • Use appropriate knowledge • See the order they exhibit • **Result:** "We appreciate not an ersatz mind but a novel kind of artefact, one whose outputs exhibit law-like regularities at the level of sign-relations, beautiful in their linguistic necessity and mechanistic grace" **FORMATTING INSTRUCTIONS:** 1. **Background**: Always #000000 (black) 2. **Section titles**: Alternate between #15E8D8, #EEFF3D, and #FC3636 3. **Slide titles**: Use colors to create visual hierarchy 4. **Emphasis**: Use #FC3636 for key points needing emphasis 5. **Block quotes**: Indent and use lighter color (#15E8D8) 6. **Bullets**: White text with colored highlights for key terms 7. **Font**: Manrope throughout 8. **Hierarchy**: Title > Main bullets > Sub-bullets This deck maintains your original structure while enriching each concept with the philosophical precision from your newer draft, particularly emphasizing the design/order distinction and the mechanical nature of LLMs throughout.