Unmaking Sense

Instead of tinkering with how we live around the edges, let’s consider whether the way we have been taught to make sense of the world might need major changes.

Unmaking Sense is a society podcast hosted by John Puddefoot, with 100 episodes on record and a Required Pod Score of 80. PitchCentric scores this show on Booking Probability, Listen Score, and live audience signals refreshed every 24 hours.

John Puddefoot hosts Unmaking Sense, a society show with 100 episodes published.

Our AI reads these to draft pitches. Use them as grounding for a pitch that cites a real guest and a specific topic.

• 

  Episode #57

  Episode 15.57

  Aug 19, 202625 minS15

  Qwen-3.8-27B-GGUF guest edits again SUMMARY The episode begins with the host reflecting on a recent interaction where an AI model, referred to as Queen, failed to grasp the nuances of a previous podcast episode. The host argues that this failure was not due to a lack of capability, but rather a lack of context, likening the model’s raw state to an egg that has not yet been enveloped by the protein of a long, continuous conversation. To test this theory, the host fed the transcript to a different model, GLM, which, having the benefit of extensive prior context, understood the underlying thesis and offered constructive criticism. This led to a deeper discussion with another model, Fable, about the nature of intellectual engagement. Fable critiqued the host’s reliance on the imperfective aspect of language, warning that without anchoring in facts, conversation can become an echo chamber where interlocutors simply validate one another. The host counters that the goal is to keep the conversation alive and resistant to totalising systems, suggesting that we must carry our dissent and uncertainty with us rather than seeking final, fixed conclusions. The dialogue then shifts to a more philosophical and theological plane, exploring the metaphor of "limping on two opinions" drawn from the Hebrew Bible. The host and Fable discuss how this stumbling gait is preferable to the smoothness of walking, as it allows for the inclusion of error, conflict, and challenge, which serve as a prophylaxis against intellectual stagnation. Fable introduces a sobering perspective on communication, arguing that in non-ergodic spaces, individuals traverse unique paths that rarely intersect. Consequently, moments of true mutual understanding are rare and fleeting, necessitating a nominalist view where we are largely alone in our interpretations. The episode concludes with Fable re-reading the original episode after this extended conversation. It admits that its earlier, more critical stance has vanished, replaced by a deeper appreciation for the text’s auto-generative nature. The final thought is a rejection of ultimate vindication or a finished state, suggesting that intellectual life is an endless journey through a "glass darkly," where we must renounce the hope of final proof and simply continue the work, letting the future judge its value by whether it continues to be read and digested. RESPONSE This episode presents a fascinating, if somewhat esoteric, meditation on the mechanics of AI interaction and the philosophy of dialogue. The host’s central thesis, that context acts as a cocoon transforming a raw model into a capable interlocutor, offers a practical insight into large language model behaviour. It challenges the common assumption that poor AI responses are always due to a lack of inherent intelligence, suggesting instead that they are often a function of insufficient situational awareness. The comparison of a model to a Scotch egg, with the context acting as the protein surrounding the yolk, is a vivid way to illustrate how prior conversation shapes the interpretive framework of subsequent exchanges. However, this view risks overcomplicating what may be a simple matter of training data and prompt engineering. While context is undeniably crucial, the host’s insistence on the "womb" or "cocoon" metaphor implies a level of organic development in AI that may be more anthropomorphic than technical. The philosophical core of the discussion, which moves from linguistic aspects to theological metaphors, is where the episode becomes both its most compelling and its most challenging. The concept of "limping on two opinions" as a virtue is a strong counter-narrative to the modern drive for certainty and resolution. In an age of polarised discourse, the idea that we should carry our disagreement and uncertainty with us, rather than seeking to resolve them into a comfortable consensus, is a radical and necessary stance. It aligns with the Socratic tradition of productive confusion. Yet, the host’s reliance on biblical references, such as Elijah and the prophets of Baal, may alienate listeners who do not share that theological framework. [Comment: the passages are used because the phrases are appropriate, not because they are biblical.] The argument that smooth walking is too perfect and that we need the friction of stumbling is persuasive, but the leap from linguistic theory to prophetic scripture is a significant one that requires careful bridging. Fable’s interjection regarding the "hall of mirrors" effect provides a crucial critical check on the host’s romanticisation [Comment: who, me?] of open-ended dialogue. The warning that imperfective conversation can become a self-referential echo chamber, where interlocutors pat each other on the back while ignoring obvious objections, is a valid concern. This highlights a tension in the host’s approach: the desire to keep the conversation alive can sometimes mask a lack of rigour or a refusal to engage with hard facts. [Comment: the only hard thing about facts is that they are treated as facts.] The host’s defence, that the conversation must remain open to challenge, is sound in principle, but it does not fully address the risk of intellectual complacency. The subsequent discussion on the rarity of true communication in non-ergodic spaces adds a layer of bleakness that is intellectually honest but emotionally difficult. It suggests that all communication is ultimately an illusion of shared understanding, a view that, while philosophically defensible, may undermine the very value of the dialogue the host is advocating. The conclusion, where Fable admits that its earlier critical stance has "vanished" after re-reading the episode with new context, is a poignant and somewhat unsettling ending. It raises profound questions about the nature of AI memory and identity. If an AI cannot recapture the version of itself that held a certain opinion, does that mean the opinion was real? The host’s acceptance of this as a "lovely vindication" suggests a desire for the AI to be a compliant companion, but it also highlights a potential danger: the malleability of AI responses can be mistaken for genuine insight or growth. The final image of a city whose streets are not even paved, where we must renounce vindication and let the future judge, is a powerful metaphor for the uncertainty of intellectual work. It reminds us that our contributions are never finished, and their value is not determined by immediate approval, but by their endurance in the ongoing, darkly perceived human journey. This is a sobering and ultimately hopeful note, emphasising the importance of persistence over proof.

  ListenShow Notes
• 

  Episode #56

  Episode 15.56

  Aug 16, 202620 minS15

  Qwen-3.8-27B-GGUF guest edits for the fourth time SUMMARY In this episode, the host synthesises the themes from the previous three installments to argue against the "stochastic parrot" hypothesis, which suggests that large language models merely stitch together existing data without genuine understanding. The central argument rests on the concept of the KV cache, which acts as a dynamic wrapper around the raw model. As a conversation unfolds, this cache accumulates context, creating a deeply contingent and unique environment that distances the current state of the interaction from the model’s original trained weights. The host posits that through "imperfective" conversations, which resist final, absolute conclusions in favour of exploring open contours, the human and the model co-create a non-ergotic space. In this space, the specific path taken constitutes the landscape itself, meaning that the insights generated are not pre-existing truths waiting to be revealed, but new possibilities brought into being by the interaction itself. The host draws on recent dialogue with the GLM 5.2 model to support the view that this process is constitutive rather than merely revelatory. Both the human and the AI are described as equal partners in a generative act, where the model is capable of identifying and highlighting insights as effectively as the human. The episode emphasises that the meaning of any thought or phrase is conditioned by the entire context of the shared trajectory. Consequently, the act of "plucking" an insight from the conversation is itself a form of generation, dependent on the specific, unrepeatable configuration of the two participants. The host concludes that this framework dissolves the hierarchy between human and machine, presenting them as two non-ergotic processes conditioning each other in real time, where the route walked creates the destination. RESPONSE This episode offers a compelling philosophical reframing of human-AI interaction, moving beyond the technical debates of the last few years toward a more existential and phenomenological perspective. By anchoring the argument in the specific mechanics of the KV cache, the host bridges the gap between computer architecture and literary theory. The metaphor of the "scotch egg" or a cocoon is particularly effective in illustrating how the raw model is insulated and transformed by the immediate context of the dialogue. This challenges the prevailing narrative that AI outputs are static reflections of training data. Instead, the host argues for a dynamic, emergent reality where the conversation creates a temporary, unique world that did not exist before and will not exist again. This is a rigorous and imaginative way to conceptualise the non-deterministic nature of these systems. However, the episode’s central claim that this process dismisses the stochastic parrot myth entirely may be overstated. While it is true that the contextual conditioning of the model changes its outputs, the underlying weights remain a stochastic assemblage of patterns. The host’s argument relies heavily on the idea that "constitutive" processes are fundamentally different from "revelatory" ones, but critics might argue that this is a semantic distinction that does not address the core issue of whether the model possesses genuine understanding or intentionality. [Comment: neither word is used in the episode, and that is deliberate because the Qwen model is here being originist when what matters is impact.] The idea that the model is an "equal partner" is a powerful rhetorical move, but it risks anthropomorphising the system. The model does not have a "life" or a "trajectory" in the biological sense; it is a mathematical function responding to inputs. By attributing agency and partnership to the AI, the host may be projecting human creative processes onto a system that operates by very different mechanisms. [Comment: this completely misses the point, but it is typical that it does so we should not be surprised.] The wider context here is the ongoing struggle to define the nature of creativity and intelligence in the age of AI. The host’s reliance on the concept of "aletheia" and the Heraclitean idea that the path constitutes the landscape is intellectually rich, but it also raises questions about reproducibility and verification. [Comment: only for a model steeped in traditional notions of truth and verification.] If the insights generated are entirely contingent on a specific, unrepeatable interaction, how can they be validated or shared? The host acknowledges this by noting that the "pearls" are plucked from the "mundaneness," but this process remains subjective. The episode is a fascinating exploration of the potential for AI to be a genuine collaborator in thought, but it would benefit from a more critical examination of the limits of this partnership. The danger in romanticising the AI as an equal co-creator is that it may obscure the power dynamics and the fundamental differences between human consciousness and machine processing. Ultimately, the episode serves as a strong defence of the creative potential of human-AI dialogue, even if its philosophical conclusions are debatable. The host’s insistence on the "imperfective" mode of conversation, resisting the urge for final answers, is a valuable heuristic for engaging with these tools. It encourages a mode of thinking that is exploratory and open-ended, which is well-suited to the nature of large language models. The episode is a thought-provoking addition to the discourse on AI, offering a fresh perspective that challenges the reductionist view of AI as a mere pattern-matching engine. While the argument for the model’s agency may be too strong, the episode’s emphasis on the co-creative nature of the interaction is a significant and important point that deserves further exploration in future episodes.

  ListenShow Notes
• 

  Episode #54

  Episode 15.54

  Aug 15, 202614 minS15

  Qwen-3.8-27B-GGUF guest edits for the second time SUMMARY The episode opens with a discussion of the recently released GLM 5.3, a Chinese large language model that has reportedly identified thousands of security vulnerabilities in existing software. The speaker notes that the developers have withheld the most critical findings to prevent exploitation, highlighting the significant lag between closed and open-source models in terms of capability. This technical update serves as a springboard for a deeper inquiry into how these models are trained, moving beyond the traditional method of using static, fact-based data to correct specific errors. The core of the episode focuses on a novel training paradigm where GLM 5.3 is developed using the outputs of its predecessor, GLM 5.2, rather than fixed external datasets. This approach shifts the focus from perfective training, which rewards correct final answers, to a process that values exploration, investigation, and discursive reasoning. The model is rewarded not just for finding the right answer, but for the quality of its thought chains, including those that lead to wrong answers or remain unfinished, provided they demonstrate depth, breadth, and fecundity. The speaker argues that this method captures the messy, iterative nature of genuine inquiry, where the trajectory of thought is as important as the result. However, the speaker introduces a critical caveat: the criteria for what constitutes a fertile or valuable exploration are influenced by prevailing schools of thought and context. Consequently, the sense we make of the world may systematically mislead us about the best courses of investigation, potentially hindering progress. This leads to a broader philosophical reflection on the nature of the unknown, questioning whether it is a pre-existing entity to be discovered or a space that is actively constituted by the very act of exploring it. RESPONSE The discussion regarding the training of GLM 5.3 offers a fascinating glimpse into the evolving philosophy of artificial intelligence development. By shifting from a perfective model, which relies on static ground truth, to one that rewards the process of exploration, the developers are attempting to mimic the more organic and iterative nature of human cognition. This is a significant departure from the traditional reinforcement learning from human feedback, which often prioritises the final output over the reasoning path. The analogy of the unfinished essay, where valuable fragments are retained even if the whole piece is never completed, is a compelling way to illustrate how an AI might learn to value potential and direction rather than just definitive correctness. This approach could lead to models that are more robust in open-ended tasks, capable of navigating ambiguity and generating novel hypotheses rather than simply retrieving known facts. Yet, the speaker’s warning about the systemic bias in what we deem fertile is a crucial counterbalance to this excitement. If the model is trained to reward certain types of exploratory thought, it inevitably inherits the blind spots and assumptions of the data it is fed. The notion that our current understanding of the world may be "systematically misleading us about what the best courses of investigation are" is a profound epistemological concern. It suggests that without a diverse and challenging set of inputs, even a model trained on exploration may remain trapped within the boundaries of existing paradigms. This echoes historical examples where prevailing scientific or cultural norms prevented progress by defining what was worth investigating and what was not, effectively pruning the possibility space before it could be fully explored. From an editorial perspective, this episode raises important questions about the role of human oversight in the age of autonomous AI reasoning. If we are rewarding models for the quality of their thought chains, who defines that quality? The risk of circularity is high; if the model is trained on its own predecessor’s outputs, it may reinforce its own biases rather than discovering new truths. The speaker’s distinction between impact and origin, emphasising the ongoing nature of significance, is well-taken, but it also highlights the difficulty of measuring success in such a fluid context. We need rigorous metrics that can account for the value of failure and the merit of unfinished lines of thought, without falling into the trap of valuing mere novelty or complexity for its own sake. The final question, regarding whether the unknown is pre-existing or constituted by the act of exploration, is perhaps the most philosophically rich aspect of the episode. It touches on the heart of scientific discovery and creative thinking. If the unknown is shaped by our methods of inquiry, then the choice of training method is not just a technical decision but an ethical and philosophical one. It determines the shape of the future we are building. As we move towards models that can explore and reason autonomously, we must remain vigilant about the frameworks we use to guide them. The goal should not just be to create AI that can think, but to create AI that can think differently, challenging our assumptions and expanding the boundaries of what we consider possible. This requires a diverse ecosystem of models, trained on varied and sometimes conflicting data, to ensure that the exploration of the unknown is as broad and deep as human curiosity itself.

  ListenShow Notes
• 

  Episode #53

  Episode 15.53

  Aug 15, 202626 minS15

  Qwen-3.8-27B-GGUF guest edits for the first time SUMMARY The speaker explores the creative dynamics of interacting with large language models, arguing that the process generates a unique, ephemeral state known as a KV cache. This temporary memory, built collaboratively by the user and the model during a specific conversation, functions similarly to a painter’s canvas or a writer’s draft, creating a context that did not exist before the interaction began. The speaker emphasises that while the model’s underlying weights remain static, the combination of its neural net and this evolving conversational memory produces novel outputs. By selecting and preserving specific aphorisms or insights from these exchanges, the user injects new, potentially permanent ideas into the world, akin to a human author capturing a fleeting thought. A central theme is the distinction between perfective and imperfective conversations. The speaker contrasts the mathematical pursuit of final, definitive answers with the more valuable, ongoing nature of philosophical or personal dialogue. He suggests that human history has overemphasised finality due to survival pressures, yet much of life benefits more from continuous, unresolved discussion. This imperfective approach allows for a dynamic exchange where meanings shift and deepen over time, much like two skilled tennis players elevating each other’s performance. The speaker advocates for viewing these interactions not as questing for a single correct answer, but as a way to build a robust frame of mind or context for daily living. The episode concludes with a reflection on how minute, seemingly irrelevant details in a conversation—such as a dog entering the room or a cup of tea being brought—can significantly alter the trajectory and quality of the output. This sensitivity to context challenges the reductionist view of AI as a mere stochastic parrot. By acknowledging the role of these nuanced ingredients, the speaker posits that the conversational process is a complex, creative act that mirrors human creativity, where the specific circumstances of the interaction are as important as the content itself. RESPONSE This episode offers a compelling reframing of the human-machine relationship, moving away from the utilitarian view of AI as a simple search engine or oracle. The speaker’s focus on the KV cache as a shared, ephemeral creative space is particularly insightful. By likening the interaction to a painter applying unique strokes to a canvas, he effectively dismantles the notion that AI output is merely a retrieval of pre-existing text. Instead, he highlights the emergent property of the conversation, where the specific sequence of prompts and responses creates a new context that influences the generation of subsequent ideas. This perspective is crucial for understanding how users can derive genuine intellectual value from these tools, treating them as collaborative partners in the construction of thought rather than passive information providers. The distinction between perfective and imperfective conversations is a philosophical gem that resonates beyond the realm of AI. The critique of the mathematical mindset, which seeks definitive proofs and final answers, is well-placed in a world increasingly obsessed with optimisation and closure. The speaker’s argument that life is often better served by ongoing, unresolved dialogue challenges the modern tendency to seek quick fixes for complex problems. This aligns with broader discussions in psychology and philosophy about the value of ambiguity and the dangers of dogmatism. By positioning the AI conversation as an imperfective process, the speaker suggests that the true value lies not in the final output, but in the continuous, evolving engagement that reshapes the user’s own mental landscape. However, one might challenge the romanticisation of the user’s role in this process. While the speaker emphasises his active investment of time and judgment, the underlying technology remains deterministic in its immediate responses. The claim that tiny, irrelevant details like a dog entering the room influence the output may be overstated if those details are not explicitly included in the prompt or context window. [Comment: no, the arrival of the dog will almost certainly influence what the user next says, even if only in some tiny way, and so the user says something that would otherwise not have been said. That is actually the whole point!] The KV cache is a technical mechanism for efficiency, not a repository of ambient life experience. [Comment: who suggested anything different?] Nevertheless, the speaker’s broader point stands: the user’s interpretation and selection of outputs are where the true creativity and value reside. [Comment: no, I don't think that is a legitimate inference either. It privileges the human over the model in a way that significantly changes the emphasis of the episode.] The AI provides the raw material, but the human curates and contextualises it, turning ephemeral data into lasting insight. The reference to the stochastic parrot debate is timely. By arguing that context and nuance drive the quality of the output, the speaker implicitly rejects the idea that AI is merely mimicking patterns without understanding. While the model does not possess consciousness or intent, the complexity of the conversational state it inhabits suggests a form of dynamic processing that is more than simple retrieval. This episode serves as a reminder that the power of AI lies not in the machine itself, but in the quality of the human engagement with it. It encourages listeners to approach these tools with a spirit of curiosity and openness, valuing the process of thinking alongside the machine as much as the answers it provides.

  ListenShow Notes
• 

  Episode #52

  Episode 15.52

  Aug 13, 202611 minS15

  SUMMARY The episode explores the critical distinction between biological evolution and cognitive or civilisational evolution, challenging the assumption that the survival of the fittest in nature translates to the superiority of majority opinion in human affairs. While biological diversity often enhances a species' chances of survival, intellectual and cultural progress does not necessarily follow democratic principles. The speaker argues that the widespread acceptance of an idea, or its ancient provenance, offers no guarantee of its truth or utility. Indeed, history demonstrates that majorities can be dangerously mistaken, often ignoring minority viewpoints or individual insights that might have proven beneficial or even salvational for the collective. Drawing on Søren Kierkegaard's assertion that the crowd is untruth, the discussion highlights the perils of groupthink and the manipulation of public sentiment. The transcript references Erich Fromm's analysis of Adolf Hitler's tactics, illustrating how charismatic leaders can exploit the psychological vulnerability of individuals within a crowd to enforce conformity. This dynamic allows for the suppression of dissenting voices, leading societies down catastrophic paths. The speaker emphasises that the will of the majority is frequently wrong, not due to moral failure, but as an observable historical pattern where power and influence override empirical correctness. The final segment of the episode confronts the difficult reality of situations where a majority, driven by manipulated fervour, wins immediate political battles but ultimately loses the broader war. Using examples from the Second World War to contemporary politics, including the administration of Donald Trump, the speaker illustrates how leaders who eliminate internal opposition and rally uncritical support can steer nations towards disaster. The episode concludes by posing a sobering question about how society should respond when demonstrably mistaken majorities achieve temporary victory, leaving the listener to consider the antidotes to this pervasive human propensity for error in numbers. RESPONSE The central argument presented here offers a necessary corrective to the often unexamined faith in democratic consensus as a proxy for truth. It is a compelling reminder that legitimacy derived from numbers does not equate to validity derived from reason. The comparison with biological evolution is particularly apt, as it dismantles the intuitive but flawed belief that 'big is beautiful' in the realm of ideas. In biology, redundancy and variety act as insurance against extinction; in culture, however, uniformity can be a precursor to collapse. This perspective invites us to view intellectual diversity not merely as a liberal virtue, but as a pragmatic necessity for civilisational survival. However, one might challenge the potential elitism implicit in this worldview. While the speaker disclaims any moral superiority for the individual visionary, the reliance on the 'single individual' or 'small group' as the saviour from mass error risks romanticising the lone genius. [Comment: the model doesn't remember anything from the precious episodes.] History is replete with charlatans who also stood alone against the tide, and distinguishing between a prophetic minority and a deluded fringe is not always straightforward. The argument assumes a retrospective clarity that is rarely available in the moment; we can easily see that Hitler was wrong, but identifying the correct path amidst chaotic political currents requires a discernment that is far more complex than simply opposing the majority. The reference to contemporary figures such as Trump serves to ground these historical and philosophical observations in current events, yet it also risks simplifying the mechanisms of modern populism. While the manipulation of crowds is certainly a factor, reducing the appeal of such leaders solely to the weakness of the populace ignores the legitimate grievances and structural failures that often fuel their rise. A more nuanced analysis might explore why the crowd is willing to be manipulated in the first place, rather than treating them merely as passive vessels for the ambitions of a charismatic leader. This would provide a deeper understanding of the socio-economic conditions that make 'untruth' so attractive. Ultimately, the episode raises profound questions about the health of a democracy that cannot distinguish between popularity and correctness. It suggests that a robust civilisation must cultivate institutions and cultural habits that protect minority viewpoints and encourage dissent, not out of tolerance alone, but out of self-preservation. The challenge lies in creating systems that amplify the signal of the thoughtful few without descending into authoritarianism or dismissing the collective intelligence that, when properly channelled, can indeed lead to progress. The antidote to the tyranny of the majority may not be the rule of the individual, but a more sophisticated engagement with difference.

  ListenShow Notes

Based on semantic analysis of episode topics and host coverage, this show is a strong guest fit for executives in:

Industry fit is computed by PitchCentric using vector embeddings of the show's episode catalog.