> [!warning] STATUS: RAW IDEAS ONLY > This file contains **brainstorming fragments and planning notes**. Nothing here is drafted. No essays are close to done. These are loose ideas to develop, not finished work. —Updated 14 Jan 2026 # Substack Ideas Dump - January 2026 A comprehensive dump of ideas for a potential philosophy/LLMs Substack. Includes article outlines, fragments, and themes. --- ## Shortlist for Launch (Refined 11 Jan 2026) Based on strategic planning sessions, these 5 articles are the priority for launch. Strategy: lead with experiential/practical pieces, save deeper philosophy for engaged audience. Skip the intro—launch *in medias res*. ### The Final 5 | # | Working Title | Core Claim | Status | |---|---------------|------------|--------| | 1 | **The Death of Texts** | Do we need books read cover-to-cover anymore? How my relationship with linear reading has changed with LLMs. | New idea | | 2 | **On Muttering** | LLMs as creativity *enablers*—extracting latent ideas from stream-of-consciousness. Counter to "AI kills creativity" narrative. | See Article 5 below | | 3 | **The Weightlessness of Authorless Text** | The phenomenology of encountering prose with no author behind it. See [[The Weightlessness of Authorless Text]]. | Has note | | 4 | **Extended Mind / Claude Code** | Working with agentic AI as genuine cognitive extension. Tie to [[Andy Clark]]'s recent work. | New idea | | 5 | **LLMs Have Made Me a Better Thinker** | Direct claim. The infinitely patient interlocutor, productive friction, cognitive traction. Reframe of Article 6. | Reframe | ### Sources for Article 4 (Extended Mind) **Andy Clark (2025):** - "Extending Minds with Generative AI" - *Nature Communications* 16:4627 - https://www.nature.com/articles/s41467-025-59906-9 - Key concept: "Digital Andy" - RAG-augmented ChatGPT trained on Clark's own work - Quote: "as human-AI collaborations become the norm, we should remind ourselves that it is our basic nature to build hybrid thinking systems" - "ChatGPT, extended: large language models and the extended mind" - *Synthese* (2025) by Smart, Clowes & Clark - https://link.springer.com/article/10.1007/s11229-025-05046-y ### Material for Article 5 (Better Thinker) **Freddie deBoer quote (antagonist position):** > "a lot of people who call themselves writers appear to spend most of their days talking themselves into believing that having ChatGPT do their work is still 'really writing.'" **My response (from [[1st Substack Article]]):** > I am one of those people. I am a philosopher: I do a lot of writing, I use LLMs to do it, and I am pretty convinced that what I am doing is really writing. Worse, I am pretty sure that I use LLMs to *think*, and that this is really thinking. ### Launch Strategy (from Jan 10 planning) 1. **Positioning:** Anti-anti-AI—not cheerleading, but a qualified rebuttal to skeptics like deBoer 2. **Cadence:** Bi-weekly, with buffer of 2-3 posts before announcing 3. **Hybrid launch:** Write 3-5 pieces, then announce specific public launch date as forcing function 4. **Freemium model:** Start mostly free to build audience 5. **Voice:** First-person intellectual honesty, willingness to be uncertain, concrete specificity --- ## Loose Ideas & Fragments ### On LLMs and Cognitive Change - NEW IDEA: I am becoming a worse typer and speaker in some ways, due to llms, should i be afraid that i am taking away might power to crystalise things? Am I? Just because what comes out of my mouth isn't that neat, doesn't mean that i am not respobsible for the ### Themes - LLMs blur the writing, reading, thinking distinction - 'a bicycle of the mind' is actually a particularly good metaphor for LLMs, a bike you can stop peddling and use momentum, an llm you can set off thinking and then pick it back up. a bike doesn't ride itself it is a dual option. - if you stop pedalling the wheels keep spinning, you keep moving, because of momentum. - and as long as you press enter, text keeps spilling across the screen ### The Absent Author - the absent author means that text can be taken on its own terms. - there's a purity to LLM text, unsullied by authorial complications. ### Potential Titles - "Well maybe I like the misery" - "AI is Good for the Soul" ### On ChatGPT's UI - worth mentioning at some point that chatgpt is primarily a textbox. and that is a weirdly interesting regression, or at least something that might look regressive, when it comes to UX/UI ### On Voice - idea for a piece on using the voice. - "a unique sensory imprint of a person" part of what makes this imprint a *unique* sensory imprint, is that it is unique to that moment. emphasise that these are your thoughts in a particular moment. then talk about how this relates to using these things with your voice. ### "Whistle While You Work" – LLMs and Mental Health - a lot about *use*, since the arrival of LLMs the way I use the computer has completely changed. - using natural language allows for *expression* even if there is nothing actually being expressed at, it certainly feels like it. - also, the way you express yourself, the turns of phrase, *do* affect the output you get back. this is satisfying. ### Promptcraft Ideas - New idea 22 Aug 2025: philosophy via prompts might seem to people like cheating, but it is actually something quite skillful and quite rewarding, promptcraft is satisfying in the way craft is satisfying (and both are at the interface and part of the environment in which they are in: natural environment, different llms. - often this is about saying how text should be augmented, changed, iterated on etc. - compare my view to JE, OB, FdB. JE – bad for thinking. Me: it doesn't feel like to me. my brain feels damn lean at the moment and i really do think that AI has done that. ### On Style - Another idea, why is style always the last thing to be done when writing with LLMs? Interesting that we still have so much trouble getting them to write right. --- ## Exhaustive Article Registry ### Article 0: An Introduction **Core Subject:** An introductory piece establishing the author's identity, the Substack's central themes, and the specific perspective from which the topic of artificial intelligence will be approached. **Author Identity:** - An analytic philosopher of mind. - Research interests: perception, agentive phenomenology, aesthetics, and generative AI. **The Substack's Dual Focus:** - The Theoretical Mode: Thinking *about* artificial intelligence as a philosopher. - The Practical Mode: Actively *using* artificial intelligence as a working philosopher. **Integrated Workflow Examples:** - Running multiple concurrent chatbots (e.g., GPT-3, Gemini 2.5 Pro, and a custom-prompted GPT-3). - Using AI for voice transcription. - Using integrated AI commands (e.g., Raycast) for on-the-fly text processing. **Caveats and Disclaimers:** - **On Technical Expertise:** Not a computer scientist; understanding is above average but below an expert (e.g., knows "gradient descent from my auto-aggressive factorization"). The blog will not focus on nuanced technical details. - **On AI Boosting:** Not an "AI booster"; aware of concerning consequences (security, environment, power concentration). These topics will not be a primary focus. Acknowledges that "this position is not without its tensions." - **On Advocacy:** Not advocating that others (e.g., artists) must embrace AI. The focus is on a personal workflow that has proven effective. The claim is that LLMs have made the author more creative, thoughtful, and a better philosopher, and sceptics can judge the work on its merits. **Personal Motivation:** - A long-standing technology enthusiast, an interest inherited from his father (e.g., MiniDisc players, computer setups). - This background led to immediate excitement upon the arrival of Midjourney and ChatGPT, a fascination that persists despite broader concerns. - Coincidentally had a large amount of free time to explore these tools when they first emerged, while waiting for a new academic job to begin. --- ### Article 1: Interrogating Texts with an LLM **Core Subject:** A practical guide to using an LLM as a research assistant for analysing academic texts. **Key Techniques (using [[Casey O'Callaghan]]'s 'Sounds' as a worked example):** - **Initial Triage:** Rapidly assessing a text's relevance. - *Example:* Ask for a one-paragraph summary, then ask a specific question like, "Is there any discussion of the relationship between sounds and their sources in this text?" - **Targeted Interrogation:** Extracting specific arguments and evidence. - *Example:* Ask for a bulleted list of reasons O'Callaghan gives for his position, then ask for a direct quote from the text that supports one of those bullet points. - **Clarification and Simplification:** Using the LLM to explain complex passages. - *Example:* After receiving a quote, prompt the model, "Explain this a little more simply." - **Argumentative Reconstruction:** Prompting the model to construct the strongest possible version of an author's argument ('steelmanning'). - *Example:* "I'd like you to *steelman* O'Callaghan's position that sounds are located in the environment, not at their sources." --- ### Article 2: Shaping LLM Cognition with System Prompts **Core Subject:** Moving beyond simple Q&A to directing an LLM's entire mode of analysis through carefully constructed initial prompts. **Key Concepts:** - Demonstrate and explain an "Ultra-Deep Thinking" prompt designed for rigorous, self-critical reasoning. - Contrast this with a "Committee of Experts" prompt that simulates a multi-perspective analysis. - Conclude by framing these system prompts as tools for specifying a *way of thinking* for the model to simulate. --- ### Article 3: Simulating an Analytic Philosophy Writing Style **Core Subject:** The difficulties of, and techniques for, getting an LLM to write in the specific style of contemporary analytic philosophy. **Structure:** - **The Problem:** Detail how generic prompts for "academic writing" lead to poor, stereotypical output. - **A Digression on a Cause:** Speculate that this is due to the training data being saturated with generic undergraduate essays. - **A Practical Solution:** Detail personal techniques, such as providing samples of one's own writing (*few-shot prompting*), creating detailed style guides, and using iterative refinement. --- ### Article 4: From Simulating Style to Simulating Thought **Core Subject:** A philosophical reflection on the implications of successfully simulating one's own writing style. **Key Philosophical Moves:** - Argue that to effectively replicate a style, the model must, to some extent, simulate the *ways of thinking* that produce that style. - Connect this back to the idea from Article 2, framing personal style simulation as the specification of a highly particular 'pattern of thought'. - Raise deeper questions about intellectual authenticity, authorship, and the relationship between style and philosophical commitment. --- ### Article 5: On Muttering and Philosophical Work **Core Subject:** Detailing a two-stage workflow using speech-to-text software for philosophical brainstorming. **The Workflow:** - **Premise:** The speed of speech allows for longer, more detailed prompts than typing. - **Stage 1: Generation (The Muttering):** A stream-of-consciousness monologue about a philosophical problem, described as "the words of a groggy middle-aged man trying to think about what he thought about yesterday, as he slowly comes to his senses in the morning." - **Stage 2: Processing:** Using an LLM to "mine the muttering for slivers of gold, or at least slivers that hold together for more than a few words." **The Expressive Dimension:** - The voice as a "unique sensory imprint" (cf. Barry Smith), a rich, analogue signal carrying epistemic markers (certainty, doubt) through prosody, tone, and tempo. - The cognitive feedback loop created by hearing one's own vocal performance. - The process as a form of embodied cognition, where the physical act of articulation is part of the thinking. **The Psychological Dynamic:** - LLMs are great at extracting meaning from text. as soon as you can give text/transcription, along with enough content, they can decipher exactly what you are getting at. people who struggle to formulate questions at conference should try this out. - The need to overcome an initial awkwardness of talking to a computer – doesn't take that long - The act of speaking to a responsive system as a method for self-clarification. **Demonstration:** Provide a "before and after" example of a raw muttering and the LLM's structured output. **The LLM dimension:** they are great at deciphering my ramblings. **Conclusion:** Reflect on the lingering cognitive role of typing, even with the physical benefits of using voice. --- ### Article 6: How LLMs Reinvigorated My Interest in Philosophy **Core Subject:** A personal reflection on how working with LLMs is more satisfying than the traditional philosophical workflow. **Contrast:** The "pain" of the pre-LLM workflow (dead ends, untrusted instincts) vs. the dynamic, interactive new method. **The LLM's Role:** An infinitely patient interlocutor. **Productive Friction:** The value of irritatedly responding to the model's "hogwash" as a method for refining one's own thoughts and the model's subsequent output. **Analogy:** The feeling of being like a sci-fi extra, able to enter a constant, productive flow ("ratatatata on the keyboard"). --- ### Article 7: Comprehension Engines & The Art of Prompting **Core Subject:** Reconceptualising LLMs as *comprehension engines* rather than knowledge engines, and exploring the philosophical skillset required for effective use. **The 'Comprehension Engine' Thesis:** Their primary utility is in helping us to better understand ideas we are already grappling with. **Use Case: The Socratic Method:** - Using a system prompt to make the LLM act as a Socratic teacher. - The goal is not to produce text, but to achieve a deep, intuitive understanding (*grokking*). **The Aesthetic Dimension (added 11 Jan 2026):** - [[Schellekens]] argues in [[Aesthetic Experience and Intellectual Pursuits]] that "coming-to-know" is genuinely aesthetic: "we come to perceive an order of things, grasp proportions, and recognize how individual parts can be made to fit into a whole." - The "felt fit" of comprehension—coherence, unity, elegance—is an aesthetic reward. - This grounds the claim that LLM-assisted understanding isn't just useful but *pleasurable* in a specifically aesthetic way. - See also: [[LLMs are understanding generators, and understanding is beautiful]] (earlier draft in _Legacy) **Philosophical Grounding:** Connect this process to formal definitions of philosophical understanding (e.g., from the paper "What is Philosophical Progress?"). **The Art of Philosophical Prompting:** - The analytic philosopher's training (clarity, precision, nuance) provides an edge in getting high-quality output. - This is the art of "stimulating LLMs, tickling their bellies." - Contrast a vague query ("What is consciousness?") with a precise, constrained philosophical prompt. - **Prompting as Debugging:** The process is a form of debugging the LLM's simulated understanding, which in turn becomes a method for debugging one's own. This raises questions of epistemic honesty vs. reinforcing bias. --- ### Article 8: Simulating and Creating **Core Subject:** An exploration of the relationship between simulation and creation in the context of LLMs. **Key Question:** Questioning the common assumption that simulation and creation are mutually exclusive. **The Argument:** Can a sufficiently complex and nuanced simulation become, for all practical purposes, indistinguishable from creation? Does the distinction ultimately matter for the philosophical utility of a novel idea generated by an LLM? --- ### Article 9: Thinking as Textual Transformation **Core Subject:** Exploring the intuition that the physical act of manipulating text between LLMs *is* the act of thinking. **The Workflow:** Characterising the process of juggling text between multiple LLMs (some with specific system prompts) as the central activity. **The Intuition:** This process replaces older methods like free-writing or discussion, and is not a record of thought but the thought itself being enacted. **Philosophical Exploration:** Connect this idea to the extended mind thesis and raise questions about where creativity resides in such a distributed process. --- ### Article 10: Being Demanding **Core Subject:** Arguing that users should be extraordinarily demanding in the tasks they assign to high-level LLMs. **The Intuition Gap:** Our sense of what constitutes a 'difficult' task is calibrated to human limitations, not machine capabilities. **Example:** Prompting an LLM to write a 4,000-word comparative analysis of two different texts—a huge task for a human, but trivial for the machine to attempt. --- ### Article 11: Content and Style **Core Subject:** Exploring how working with LLMs challenges the traditional distinction between the content of a text and its style. **The Argument:** Desirable textual qualities (like clarity, rigour, or conciseness) can be treated as operational parameters in a generative process. Prompting a model to "increase the rigour" blurs the line between changing a text's substance and altering its form. --- ### Article 12: The Satisfaction of Active Engagement **Core Subject:** Examining the affective and phenomenological character of doing philosophy with LLMs. **The Contrast:** The traditional, passive, contemplative mode of 'reading and pondering' vs. the active, generative, productive mode of working with an LLM. **The Source of Satisfaction:** The instantaneous feedback loop and constant sense of forward momentum in the LLM workflow. --- ### Article 13: The Division of Labour **Core Subject:** Exploring the tension between productive delegation and intellectual laziness when using LLMs. **The Dilemma:** The need to "train the muscle that is thinking" vs. the efficiency of offloading cognitive effort for trivial problems or verifying the "fuzzy edges" of one's thinking. **The Goal:** To develop a personal policy for dividing intellectual labour, discerning when effort is formative and when it is merely inefficient. Strategic laziness as a new skill.