Pierre Beckmann

The individuation problem for large language models asks which entities associated with them, if any, should be identified as minds. We approach this problem through mechanistic interpretability, engaging in particular with recent empirical work on persona vectors, persona space, and emergent misalignment. We argue that three views are the strongest candidates: the virtual instance view and two new views we introduce, the (virtual) instance-persona view and the model-persona view. First, we argue for the virtual instance view on the grounds that attention streams sustain quasi-psychological connections across token-time. Then we present the persona literature, organised around three hypotheses about the internal structure underlying personas in LLMs, and show that the two persona-based views are promising alternatives.

Recent work in mechanistic interpretability has studied how large language models recall facts stored in their weights. This paper argues that factual recall points to something broader: a general kind of operation in deep learning models, which I call feature recall. The core observation is that a linear projection can be read as retrieving stored information scaled by input activations. I define feature recall, show it applies across architectures, and contrast it with the established paradigm of feature combination. I also consider how cases of feature recall might be mechanistically identified. The account offers philosophers a new conceptual tool for understanding deep learning and points to empirical research directions for mechanistic interpretability research.

OpenAI has recently released SORA, a deep learning model that can generate highly realistic videos. Its creators claim that it “understands the physical world in motion.” In this paper, I subject this claim to philosophical scrutiny. After explaining in general how stable diffusion models generate videos, I employ the concepts of explanatory and objectual understanding to determine what kind of understanding of the physical world such deep learning models for video generation might possess. Drawing on recent literature in both epistemology and the philosophy of science, I build a set of conditions under which such kinds of understanding might be attributed to SORA and to deep learning models in general. This allows me to spell out the sense in which these models may be said to understand the world, and to uncover the primary axes for evaluating the degree of such understanding. My key finding is that consistency, both across outputs and with underlying operations, is crucial when attributing understanding to deep learning models.

What makes understanding an important cognitive state? And what does having the concept of understanding do for us? This paper offers a unifying account of understanding by jointly reverse-engineering the function of both the state and the concept. We argue that we care about understanding because it grounds and predicts robust competence: the stable ability to succeed across novel scenarios. Our concept of understanding evolved as an efficient proxy to track this elusive property, allowing us to identify who to trust and learn from. This highlights the sociality of understanding and how it shapes what kinds of understanding we are apt to form. We then argue that understanding is the result of convergent pressures on social agents to predict the world using models that are not only accurate, but also compressed enough to be stored, demonstrated, and transmitted. This allows us to integrate a number of ostensibly competing accounts of understanding. Finally, we show how the forces at the root of human understanding elucidate debates over AI understanding.

We propose a non-representationalist framework for deep learning relying on a novel method computational phenomenology, a dialogue between the first-person perspective (relying on phenomenology) and the mechanisms of computational models. We thereby propose an alternative to the modern cognitivist interpretation of deep learning, according to which artificial neural networks encode representations of external entities. This interpretation mainly relies on neuro-representationalism, a position that combines a strong ontological commitment towards scientific theoretical entities and the idea that the brain operates on symbolic representations of these entities. We proceed as follows: after offering a review of cognitivism and neuro-representationalism in the field of deep learning, we first elaborate a phenomenological critique of these positions; we then sketch out computational phenomenology and distinguish it from existing alternatives; finally we apply this new method to deep learning models trained on specific tasks, in order to formulate a conceptual framework of deep-learning, that allows one to think of artificial neural networks’ mechanisms in terms of lived experience.

Large language models (LLMs) are often portrayed as merely imitating linguistic patterns without genuine understanding. We argue that recent findings in mechanistic interpretability (MI), the emerging field probing the inner workings of LLMs, render this picture increasingly untenable—but only once those findings are integrated within a theoretical account of understanding. We propose a tiered framework for thinking about understanding in LLMs and use it to synthesize the most relevant findings to date. The framework distinguishes three hierarchical varieties of understanding, each tied to a corresponding level of computational organization: conceptual understanding emerges when a model forms “features” as directions in latent space, learning connections between diverse manifestations of a single entity or property; state-of-the-world understanding emerges when a model learns contingent factual connections between features and dynamically tracks changes in the world; principled understanding emerges when a model ceases to rely on memorized facts and discovers a compact “circuit” connecting these facts. Across these tiers, MI uncovers internal organizations that can underwrite understanding-like unification. However, these also diverge from human cognition in their parallel exploitation of heterogeneous mechanisms. Fusing philosophical theory with mechanistic evidence thus allows us to transcend binary debates over whether AI understands, paving the way for a comparative, mechanistically grounded epistemology that explores how AI understanding aligns with—and diverges from—our own.