# 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 ``` # conversation with Egor Kraft blockchain image justification name of organisation/decentralised software project hashd0x https://hashdox.org/ https://kraft.studio/hashd0x-proof-of-war/ https://proofofwar.hashdox.org/ # Venice Conference day three #veniceconference 'the camera looks out in the world but the model looks in on itself' ## aesthetics of genericity talk point 4 why aren't AI images 'even images'? # first draft of 1000-word extended abstract for [[Generating Philosophy Kanban]] paper #paper/generatingphilosophy > "Forty-two," said Deep Thought, with infinite majesty and calm. > It was a long time before anyone spoke. > Out of the corner of his eye Phouchg could see the sea of tense expectant faces down in the square outside. > "We're going to get lynched aren't we?" he whispered. > "It was a tough assignment," said Deep Thought mildly. > "Forty-two\!" yelled Loonquawl. "Is that all you've got to show for seven and a half million years' work?" > "I checked it very thoroughly," said [[the computer]], "and that quite definitely is the answer. I think [[the problem]], to be quite honest with you, is that you've never actually known what [[the question]] is." > \-- Douglas Adams, *The Hitchhiker's Guide to the Galaxy* In The Hitchhiker's Guide to the Galaxy, humanity asks an AI to do some philosophy; a computer named Deep Thought is constructed and instructed to provide "The Answer to the Ultimate Question of Life, the Universe, and Everything." Humanity receives the answer '42' – an answer which, while apparently correct, means next to nothing at all due to humanity’s failure to [[know what]] the Ultimate Question in fact is. In 2025, we are in a position to think about [[the relationship between]] philosophy and AI for real. Can LLMs enhance [[philosophical understanding]]? They are happy to dispense philosophical wisdom if we ask them to, but should we listen? Many might doubt that we should, given that LLMs are notoriously prone to ‘hallucinations’. In this paper, I argue for cautious optimism: the creators of such systems sometimes describe them as *reasoning engines*, and I suggest this description is broadly accurate. Although LLMs themselves do not reason in the way that people do, they can simulate human reasoning, and through this simulation, they are capable of enhancing the [[philosophical understanding]] of their users. To assess how LLMs might enhance [[philosophical understanding]], we first require a definition of this understanding. In this paper, we adopt the framework proposed by Dellsén et al. (2024). Their approach, explaining understanding in terms of [[representing dependence networks]], resonates with the broader philosophical aim, articulated by Sellars (1962, p. 1), to grasp 'how things in the broadest possible sense of the term hang together in the broadest possible sense of the term'. [[The central claim]] holds that the degree to which a subject understands a phenomenon corresponds to the accuracy and comprehensiveness of their representation of the network of [[dependence relations]] relevant to that phenomenon. These [[dependence relations]] are conceived as objective, worldly relations often underpinning explanations – candidates include constitution, grounding, and supervenience, amongst others. Understanding, on [[this view]], admits of degrees relative to accuracy and comprehensiveness. This account is described as "robustly factive," requiring the representation to correspond to actual dependencies, yet also "epistemically undemanding" (Dellsén et al. 2024, pp. 674, 676). Understanding X, in this specific sense, does not require having justification for, or even belief in, the propositions represented. This specific notion – understanding as accurate and comprehensive representation of dependence networks, without requiring justification – guides the subsequent analysis. The potential for LLMs to contribute to this refinement can be considered. LLMs are proficient at answering certain low-level, fact-based questions, and this can enhance the comprehensiveness of a user's representation of [[dependency relations]]. While it might be objected that LLMs ‘hallucinate’ and so should not be trusted for even these questions, they often excel at well-worn facts. A similar capacity is restructuring text, including simplification or providing analogies. These may be basic enhancements, but they are enhancements nonetheless. A more central question, of course, is the extent to which LLMs can produce novel outputs which enhance their users’ philosophical understanding. An initial objection one might have is that these systems do not themselves understand anything at all; they are ‘stochastic parrots’, incapable of doing anything other than arranging words in statistically plausible patterns. This may well be true; however, we should not assume that they need to reason in order to produce philosophically illuminating outputs. Butlin and Viebahn (forthcoming) suggest that fine-tuned LLMs may produce outputs with a ‘descriptive function’, defined as "the function of conveying information to an observer or consumer system, so as to cause the consumer to behave as though some condition holds" (p. 4). Fine-tuning enhances the reliability of the information communicated, equipping LLM outputs with this descriptive function, analogous to the way a thermometer is designed to indicate temperature. Simply functioning to convey a slice of information, however, does not seem enough to enhance someone’s understanding. Consider again Deep Thought: what makes the answer ‘42’ so unsatisfying is its failure to illuminate any relevant dependence relations. More broadly, philosophical understanding is enhanced through exposure to argument. Our question then becomes: can LLMs produce philosophical arguments? LLMs are prone to errors, sometimes on tasks that seem straightforward analytically but fall outside their core competency of generating plausible text, such as the 'strawberry problem' where models incorrectly count letters due to tokenisation processes and a training objective focused on next-token prediction. Perhaps surprisingly, models demonstrated considerably improved performance on this sort of analytical problem when prompted with instructions such as ‘Let’s think step by step’. This observation spurred the development of specific guidance techniques, most notably Chain-of-Thought (CoT) prompting. CoT prompting involves instructing the model – either by providing explicit examples of step-by-step reasoning or by using direct textual commands – to articulate intermediate stages en route to formulating a final answer. It must be emphasised that this process constitutes a simulation of reasoning. The model’s capacity to produce coherent reasoning chains stems not from manipulating internal logical representations or possessing genuine conceptual grasp, but rather from having learned complex statistical patterns embedded within its vast training data, which includes innumerable examples of human arguments and explanations. CoT prompting essentially activates these learned patterns. The quality and reliability of this simulation are known to depend significantly on factors such as model scale and refinement techniques like Instruction Fine-Tuning (IFT) and Reinforcement Learning from Human Feedback (RLHF). Despite its nature as simulation, this capability demonstrates effectiveness in practice; CoT-guided models show improved performance across an array of benchmarks requiring multi-step reasoning. This effectiveness connects directly to the potential for enhancing philosophical understanding. The simulation of reasoning allows LLMs to generate outputs structured much like arguments, presenting sequences of claims apparently linked by inferential markers. The philosophical utility derived from engaging with these simulated arguments mirrors our engagement with conventional, human-authored philosophical texts. A philosophical argument need not be perfectly sound for it to be of use; the process of engaging with an argument prompts the reader to critically examine purported dependencies. The fact that the reasoning process is simulated does not preclude the resulting textual artifact from serving as a stimulus for philosophical reflection. Consequently, the structured outputs generated via CoT prompting can directly enhance philosophical understanding as defined by the Dellsén framework. The utility resides not in attributing comprehension or insight to the LLM itself, but rather in how the generated text, by virtue of its structure, can prompt specific kinds of refinement in the user's own representational framework. First, by presenting a step-by-step derivation, CoT outputs enable the user to meticulously scrutinise the purported dependence relations involved. Identifying a weak or invalid link prompts the user to correct their own mental model. Second, the intermediate steps articulated within a CoT output can introduce new nodes or relations into the user's existing dependence network, thereby increasing its comprehensiveness. The model might highlight a subtle distinction or surface an implicit premise. Third, the step-by-step structure inherent in CoT outputs can aid in clarifying the nature or type of the proposed dependencies. This aligns with the ‘epistemically undemanding’ nature of the Dellsén et al. (2024) framework. Enhancing understanding, on this view, does not strictly require the user to possess justification for every proposition contained within the LLM's generated output. The primary utility derives from the potential of the generated structure itself to provoke a critical re-evaluation and subsequent refinement of the user's own existing representation. In sum, while these systems lack genuine understanding, their ability to generate structured, step-by-step textual artifacts through techniques like CoT is pertinent. These artifacts, when engaged with critically by a human user, can directly contribute to enhancing philosophical understanding – as defined by the accurate, comprehensive, and clear representation of dependence networks – by facilitating the scrutiny, correction, and expansion of the user's own mental models. LLMs can thus function as tools for philosophical inquiry, providing artefacts whose analysis can contribute to understanding in the user, irrespective of the non-understanding nature of their source. Philosophers might leverage these outputs as starting points for analysis, as mechanisms for exploring conceptual connections, or as generators of alternative perspectives. What appears pertinent is not the cognitive status of the LLM, but rather the capacity of its structured output to prompt and inform human philosophical thought. **