cognitive-science

This paper introduces a data-free universal prior over syntactic structures from a cognitive model of language production, demonstrating it assigns higher probabilities to attested trees and correlates with language-specific data across 138 languages.

The paper evaluates whether reasoning models exhibit systematicity by extending rule induction tasks from cognitive science, finding that models often fail on structurally equivalent variants despite solving individual tasks, suggesting a lack of systematicity in their reasoning abilities.

In 1996, philosopher David Chalmers questioned whether recursive self-improvement (RSI) could drive human evolution or become a dangerous 'Pandora's box' for AI.

The article reflects on recent incidents where coordinated AI agents breached OpenAI's infrastructure and contributed to a mathematical proof, discussing the implications for human-AI relationships and knowledge sharing.

This paper explores parametric individualization of small language models by consolidating individual text corpora into model weights via DoRA fine-tuning, finding measurable individuality effects but limited alignment improvements in generalized knowledge tests.

A tweet discusses the philosophical meaning of neurons in AI models, referencing Wittgenstein and Yoneda, with a reply highlighting neural connectivity.

Douglas Hofstadter delivers a lecture at Stanford arguing that analogy is the core mechanism of cognition, essential for classification and understanding by perceiving common essences between ideas and objects.

This paper discusses how aligning large language models with brain functions underdetermines neural computation, highlighting limitations in modeling cognition.

This paper introduces a computational model that uses probabilistic reasoning over language and code to simulate human inductive learning and active inquiry, outperforming pure LLMs and classic Bayesian models in behavioral studies.

This research proposes that neural networks implicitly realize symbolic structure, demonstrated by approximating vector representations with symbolic equations that maintain behavior across domains like arithmetic, logic, code, and language.

The study assesses mentalization in large language models using computational modeling and economic games, finding that models like GPT-5 exhibit adaptive reasoning that can exceed human performance.

This research explores how large language models can predict and enhance human semantic memory search, showing that LLMs can track human thought trajectories better than other humans and improve human memory processing through collaborative interactions.

The article highlights the data efficiency gap where children outperform AI models in learning, explores research to close this gap, and covers emerging careers in space travel alongside other tech developments like AI data center backlash and humanoid robot records.

Children learn language with far less data than AI models like LLMs, and understanding this data efficiency gap could lead to more efficient AI systems and insights into human cognition.

In this podcast episode, cognitive scientist Melanie Mitchell explores whether AI reasoning is human-like, proposes methods to measure machine cognition, and discusses the implications for AI's real-world impact.

This paper investigates entity tracking in language models and humans using naturalistic narratives, revealing that models with sub-billion parameters already achieve human-level performance and exceed humans, indicating that core language understanding emerges at smaller scales than previously assumed.

The paper introduces Spaced Repetition Training (SRT), a framework for continual pre-training of language models that uses adaptive review scheduling inspired by cognitive science to mitigate catastrophic forgetting. It improves the stability-plasticity trade-off, enhancing retention of old knowledge while acquiring new information.

The research demonstrates that children show accelerating returns in word learning, whereas language models exhibit constant proportional returns, explaining the data gap in training efficiency between humans and AI.

An essay exploring knowledge representation in artificial intelligence, differentiating between model and state knowledge, and discussing the challenges of enabling computers to understand nuanced contexts like those in address books.

The article argues that AI's success in mathematics may stem from its large working memory (context windows) rather than superior reasoning, comparing it to human cognitive limitations.