# 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 ``` # Presentation Planning #veniceconference ## Introduction **Central Claim** _Critics should take up prompting._ ### Naïve criticism of prompting as art - 'Prompting is nothing but instruction‑giving, it is not an artistic skill like painting, sculpting, composing' - A prompt reads like a _rule set_: “style of Monet”, “high‑key lighting”, “ISO 800 film grain”. - The user supplies labels; the model executes—no brushwork, no phrasing, no performance. - From this angle, prompting is closer to writing a command line than to composing or painting. - **Execution is automatic** - The pretrained network carries out the substantive aesthetic labour—composition, colour harmony, texturing. - Whatever beauty appears is attributable to correlations internalised from vast datasets, not to [[the prompter]]’s hand. - **No material resistance** - Oils congeal, clay slumps, marble chips; those frictions teach craft. - Keyboard input meets no analogue drag; every iteration costs only a few seconds of GPU time. - **Sparse improvisation** - Traditional makers adjust stroke by stroke, note by note. - Prompting unfolds in discrete cycles: type → generate → wait, without tactile micro‑adjustment. - **Credit accrues to the model, not the user** - The architecture and its [[training corpus]] embed the "style"; the user merely cherry‑picks it. - [[The prompter]] therefore [[looks like]] an operator selecting presets, not an artist shaping matter. - **Up‑shot of the naïve view** - Prompting = issuing rules to a sophisticated copier. - It ranks with telling a studio assistant, “Paint my portrait in the style of Vermeer”—efficient, but not creative. \ #### Plan for the 20‑minute talk - **Step 1 – Sibley on critical perception** Lay out Sibley’s point that aesthetic judgement depends on _seeing_ qualities such as grace or balance, not on inferring them, and that the critic’s job is to help others _see for themselves_—what he calls “perceptual proof”. - **Step 2 – Carry those skills into prompting** Show that a prompter must use the same eye to recognise when an output carries the intended quality and must rely on the same descriptive vocabulary to steer the model. - **Step 3 – Show [[the key difference]]** Critics perform _speech acts_—they talk to people so that those people see what they see. Prompting is not a speech act at all; it is more like turning knobs on a machine. The words are control signals sent to a model, not sentences addressed to an audience. Describe prompting as trial‑and‑adjustment with a [[generative system]] whose **[[dynamic recalcitrance]]** resists simple instruction; success comes from iterative guidance, not fixed rules. - **Step 4 – Pay‑off** Since critics already have the needed eye and words, they are the group best placed to adopt prompting as a natural extension of their practice. --- ### Critics and Prompters: The Common Toolkit This note sets out, **only in the positive**, the two capacities that Sibley’s critic must possess and shows how each capacity is required—unchanged in kind—by an effective AI prompter. Comparisons, objections, and differences are postponed. --- #### 1. Refined Perceptual Sensitivity |Role|What must be perceived|Why it is indispensable| |---|---|---| |**Critic**|Emergent aesthetic qualities in an extant work (e.g. grace, balance, tension).|Without first _seeing_ the quality, the critic has nothing to explain or share.| |**Prompter**|The emergence (or absence) of the aimed‑at quality in each generated image.|Without recognising elegance, unity, mood, etc., iteration becomes blind trial‑and‑error; [[the prompter]] would not know when to stop or how to adjust.| _Summary:_ Both roles are idle unless the agent can directly detect aesthetic qualities. --- #### 2. Descriptive & Directive Vocabulary |Role|Linguistic stock‑in‑trade|Function of the language| |---|---|---| |**Critic**|Style labels, colour and value terms, compositional descriptors, historical references, metaphors.|Direct a viewer’s attention to responsible features; render perception articulate.| |**Prompter**|The _same_ lexicon—"oil‑study chiaroscuro", "muted tertiary palette", "high‑key geometric abstraction", camera models, film stocks, art‑historical styles.|Serve as prompt tokens that steer the model towards regions of [[latent space]] statistically correlated with those features.| _Summary:_ The critic’s vocabulary for pointing‑out becomes [[the prompter]]’s vocabulary for pointing‑in. --- #### Interim Thesis > **A successful AI prompter needs precisely the two capacities Sibley assigns to the competent critic:** > **(i) a trained eye that discerns emergent aesthetic qualities, and** > **(ii) an articulated, feature‑rich vocabulary capable of directing aesthetic attention—whether that attention belongs to another human mind (criticism) or to the statistical machinery of a generative model (prompting).** The sections that follow flesh out Steps 1–3 in turn. --- ### Sibley on critical perception - **Seeing, not deducing** - “People have to _see_ the grace or unity of a work” – sight, hearing, or felt response is the entry ticket to any aesthetic judgement. - Following rules or applying criteria can label a work, but unless the quality is _seen_, no genuine judgement occurs. - **Emergent aesthetic qualities** - Balance, grace, tension, gaiety: none are reducible to line‑length, colour‑mix, or tempo alone. - They _depend_ on non‑aesthetic features yet outstrip them; a small change in a line’s curve can destroy grace. - **Critic’s double assignment** 1. **Notice first** – The critic must perceive these emergent qualities unaided. 2. **Make others notice** – Through words, comparisons, metaphors, gestures, the critic “shows” rather than “proves”. Sibley calls this _perceptual proof_. - He stresses that such showing is _not_ giving reasons that logically guarantee the judgement; success is the listener _seeing for themselves_. - **No rule‑substitution** - Because no set of non‑aesthetic descriptions is sufficient, rules cannot replace the critic’s trained eye. - Hence criticism is irreducibly perceptual and pedagogical. ### Critics and Prompters: Shared Toolkit (expanded) > Sibley’s vocabulary underwrites every item below: _emergent qualities_ must be **seen**, and language must _direct attention_ to the non‑aesthetic “responsible features.” The same logic governs prompt‑craft. - **Capacity 1 – Trained eye (perceptual sensitivity)** - _Critic_: perceives grace, imbalance, tonal disunity _directly_; these are Sibley’s _aesthetic qualities_ that cannot be deduced. - _Prompter_: inspects each output for those very emergent qualities; only perception signals when the loop should terminate. - _Sibley link_: the stopping‑criterion is a private act of _perceptual judgement_—[[the prompter]] mirrors the critic’s “notice first.” - _Further sub‑point_: If elegance is missing, no amount of extra tokens will suffice—a live echo of Sibley’s “no rule‑substitution.” - **Capacity 2 – Feature‑rich vocabulary (pointing language)** - _Critic_: “cool tertiary palette,” “dynamic diagonal,” “syncopated metre” — phrases that single out **non‑aesthetic features** Sibley treats as _responsible_ for the emergent property. - _Prompter_: re‑deploys the same phrases; the model aligns colour distributions or compositional templates accordingly. - _Sibley link_: [[the prompt]] acts like the critic’s **gesture** or **metaphor**—it directs (here statistically) toward the features that give rise to the quality. - _Further sub‑point_: Prompt tokens inherit their force from art‑historical usage; this preserves the critic’s pedagogical lexicon. - **Capacity 3 – Iterative judgement (perceptual proof in reverse)** - _Critic_: may revisit a work, refining description until the listener _sees_—Sibley’s _perceptual proof_. - _Prompter_: prompt → output → tweak → output—the refinement continues until _she_ sees; it is perceptual proof turned inward. - _Sibley link_: each cycle re‑tests the hypothesis that certain non‑aesthetic tweaks will _emit_ the target quality; failure re‑confirms Sibley’s point that rules never compel success. - **Interim takeaway** - The critic’s toolkit transfers intact; the key shift is the _addressee_: human mind ↔ generative model. Sibley’s insistence on perception and responsible features still frames the activity. ### Prompting: control signals, not [[speech acts]] |Dimension|Criticism (speech act)|Prompting (control signal)| |---|---|---| |**Addressee**|Another person with perception and judgement.|A generative model that parses tokens into numbers.| |**Function of words**|Draw attention; foster shared seeing.|Shift probability mass; nudge latent variables.| |**Feedback loop**|Immediate verbal or visual cues from the listener.|Stochastic image output; user inspects then re‑prompts.| |**Type of resistance**|Audience incomprehension; interpretive blind‑spots.|Model bias, randomness, training artefacts.| |**Measure of success**|Audience now _sees_ the quality unaided.|User finally _sees_ the desired quality in the output.| |**Credit for outcome**|Shared between artist and critic’s guidance.|Shared between dataset/architecture and user’s steering.| - **Words address machines, not minds** A prompt is converted into vectors; the model has no comprehension, only parameter adjustment. - **Negotiation, not instruction list** Naïve picture: prompt = rule list → fixed image. Reality: the model’s _dynamic recalcitrance_ (noise seeds, sampling temperature, training bias) forces trial‑and‑error. - The user treats each output as a probe, then trims or expands [[the prompt]] accordingly. - This loop mirrors a craftworker testing and correcting each stroke. - **Philosophical up‑shot** The move from speech to control shifts prompting away from hermeneutics and towards _operative craft_. What remains constant is [[the need]] for perceptual judgement to declare an iteration “finished”. --- # New version of AI presentation plan #veniceconference PLAN: ## Central Claim - Critics should take up prompting. ## The Naïve View: Prompting as Mere Instruction, Not Art - An initial reaction to AI image generation is to view the act of prompting not as an artistic practice, but as a form of instruction-giving, akin to providing a recipe or a set of commands. - This perspective typically rests on several observations: - **Prompts as Rule-Sets:** - From this viewpoint, a prompt appears to function like a list of explicit instructions or rules. - Phrases such as “style of Monet,” “high‑key lighting,” or “ISO 800 film grain” are interpreted as commands given to the AI system. - The user supplies descriptive labels, and the model is then expected to execute these directives. - There is no apparent equivalent to the painter's brushwork, the writer's phrasing, or the musician's performance in the act of typing these textual commands. - **Automated Execution of Aesthetic Labour:** - A point supporting this view is that the pretrained neural network seems to carry out the aesthetic work. - Tasks of composition, colour harmonisation, and the application of textures appear to be handled by the model itself. - Any resulting beauty or aesthetic appeal is then attributed to the statistical correlations and patterns the model has internalised from its vast training dataset, rather than to any artistic skill demonstrated by the prompter during the act of generation. - The human's role, in this light, is to specify _what_ to generate, not to engage in the _how_ of its aesthetic realisation. - **Apparent Lack of Engagement with Media Recalcitrance:** - Traditional artistic practices are often characterised by an engagement with the _recalcitrance_ of a chosen medium. Recalcitrance, in this sense, refers to the difficulties, resistances, and challenges that materials and tools present, which the artist must grapple with and respond to in the process of creation (cf. Wollheim 1980; Young & Terrone 2024). - For instance, oils congeal and have specific drying times, clay slumps under its own weight if not properly handled, and marble chips unpredictably, demanding careful force. These material frictions are not obstacles but are often part of the development of embodied craft and skill; the artist learns through this negotiation with the medium's inherent properties. - From the naïve viewpoint, prompting as instruction-giving appears to circumvent this engagement with recalcitrance. Keyboard input, when seen as issuing commands, does not seem to encounter material push-back. - If the AI model executes instructions, the process might seem to bypass the negotiation with a resistant medium that is part of many traditional crafts. The "cost" of iteration being primarily computational (a few seconds of GPU time) rather than involving physical effort or engagement with material properties further reinforces this perception of diminished recalcitrance. Telling a system _what_ to do seems to involve little of the struggle with a resistant entity that defines much artistic labour. # Critics and Prompters: Shared Capacities (Reinforcing the Naïve View?) - Furthermore, the view that prompting is instruction-giving, rather than artistic practice, might seem strengthened when we consider the capacities required for effective prompting. These capacities appear to mirror those established for the art critic. - Sibley’s account underwrites the items below: emergent qualities must be seen, and language must direct attention to the non‑aesthetic “responsible features.” The apparent fact that the same logic governs prompt‑craft could suggest that prompting is an extension of critical description, not artistic creation. - **Capacity 1 – Trained eye (perceptual sensitivity)** - Critic: perceives grace, imbalance, tonal disunity directly; these are Sibley’s aesthetic qualities that cannot be deduced. - Prompter: inspects each output for those very emergent qualities; only perception signals when the loop should terminate. - Sibley link: the stopping‑criterion is a private act of perceptual judgement—the prompter mirrors the critic’s “notice first.” - Further sub‑point: If elegance is missing, no amount of extra tokens will suffice—a live echo of Sibley’s “no rule‑substitution.” - **Capacity 2 – Feature‑rich vocabulary (pointing language)** - Critic: “cool tertiary palette,” “dynamic diagonal,” “syncopated metre” — phrases that single out non‑aesthetic features Sibley treats as responsible for the emergent property. - Prompter: re‑deploys the same phrases; the model aligns colour distributions or compositional templates accordingly. - Sibley link: the prompt acts like the critic’s gesture or metaphor—it directs (here statistically) toward the features that give rise to the quality. - Further sub‑point: Prompt tokens inherit their force from art‑historical usage; this preserves the critic’s pedagogical lexicon. - **Capacity 3 – Iterative judgement (perceptual proof in reverse)** - Critic: may revisit a work, refining description until the listener sees—Sibley’s perceptual proof. - Prompter: prompt → output → tweak → output—the refinement continues until she sees; it is perceptual proof turned inward. - Sibley link: each cycle re‑tests the hypothesis that certain non‑aesthetic tweaks will emit the target quality; failure re‑confirms Sibley’s point that rules never compel success. ## Interim takeaway (from the Naïve perspective) - The capacities required for criticism appear to transfer to prompting. If the prompter primarily uses the critic's perceptual skills and descriptive language, this might reinforce the view that prompting is about describing and directing, rather than creating. # Sibley on critical perception - **Seeing, not deducing** - “People have to see the grace or unity of a work” – sight, hearing, or felt response is the entry ticket to any aesthetic judgement. - Following rules or applying criteria can label a work, but unless the quality is seen, no aesthetic judgement occurs. - **Emergent aesthetic qualities** - Balance, grace, tension, gaiety: none are reducible to line‑length, colour‑mix, or tempo alone. - They depend on non‑aesthetic features yet outstrip them; a small change in a line’s curve can destroy grace. - **Critic’s double assignment** - Notice first – The critic must perceive these emergent qualities unaided. - Make others notice – Through words, comparisons, metaphors, gestures, the critic “shows” rather than “proves”. Sibley calls this perceptual proof. - He stresses that such showing is not giving reasons that logically guarantee the judgement; success is the listener seeing for themselves. - **No rule‑substitution** - Because no set of non‑aesthetic descriptions is sufficient, rules cannot replace the critic’s trained eye. - Hence criticism is irreducibly perceptual and pedagogical. # Prompting: control signals, not speech acts - One might think that connecting prompting and criticism, particularly the shared use of a descriptive vocabulary (as outlined in the previous section), reinforces the idea that prompting is a form of instruction-giving. If the prompter uses the critic's words, perhaps better vocabulary just leads to better instructions. - However, this overlooks a distinction in the _nature_ of the acts performed. Just because the prompter uses the same words as a critic, prompting and criticism are not the same kind of activity. The difference lies in whether the words function as part of a _communicative_ act directed at a comprehending mind, or as _control signals_ directed at a non-comprehending mechanism. - Criticism, in Sibley's model, is a speech act. The critic uses words (descriptions, metaphors, comparisons) to address another person who possesses perception and judgement. The function of these words is communicative: to draw the listener's attention to certain features and thereby foster a shared perceptual experience or understanding ("See the grace _here_"). - Prompting, by contrast, is not a communicative act in this sense. The addressee is not another person capable of shared perception or understanding, but a generative model – a system that parses linguistic tokens into numerical representations (vectors). The function of the prompt words is not to foster understanding in the model, but to act as control signals: to adjust parameters, shift probability distributions within the model's latent space, and nudge the generative process towards certain statistical correlations associated with those tokens. The model does not 'understand' the prompt; it reacts to it computationally. - **Dimension: Feedback loop** - _Criticism (speech act):_ Verbal or visual cues from the listener. - _Prompting (control signal):_ Stochastic image output; user inspects then re‑prompts. - **Dimension: Type of resistance** - _Criticism (speech act):_ Audience incomprehension; interpretive blind‑spots. - _Prompting (control signal):_ Model bias, randomness, training artefacts. - **Dimension: Measure of success** - _Criticism (speech act):_ Audience now sees the quality unaided. - _Prompting (control signal):_ User finally sees the desired quality in the output. - **Dimension: Credit for outcome** - _Criticism (speech act):_ Shared between artist and critic’s guidance. - _Prompting (control signal):_ Shared between dataset/architecture and user’s steering. - **Words address machines, not minds** - A prompt is converted into vectors; the model has no comprehension, only parameter adjustment. - **Negotiation, not instruction list** - Naïve picture: prompt = rule list → fixed image. - Reality: the model’s dynamic recalcitrance (noise seeds, sampling temperature, training bias) forces trial‑and‑error. - The user treats each output as a probe, then trims or expands the prompt accordingly. - This loop mirrors a craftworker testing and correcting each stroke. ## Philosophical up‑shot - The move from speech to control shifts prompting away from hermeneutics and towards operative craft. - What remains constant is the need for perceptual judgement to declare an iteration “finished”. # the many hands problem of generative ai - 'who is the creator'? can't we say this with so many other types of art - relevance and non-redundancy –why should we consider the - control 'stable diffusion has control' - originality action painting Upstream co-creators, is that really going to be as simple as in the Jana case? I can make