Semantic geometry emerges transiently early in next-token prediction training before collapsing to Neural Collapse symmetry in synthetic settings with latent semantic factors.
arXiv preprint arXiv:2306.11932 , year=
10 Pith papers cite this work, alongside 20 external citations. Polarity classification is still indexing.
10
Pith papers citing it
20
external citations · external index
citation-role summary
background 1
dataset 1
citation-polarity summary
representative citing papers
Introduces the plurality matrix and a level-based stratification of disagreement measures, proving many require triple-wise information and providing two elicitation protocols that trade off participant numbers against cognitive load.
Synthetic training data designed to break the correlation between semantic and preferential signals in text embeddings provably improves preference prediction across 11 online deliberation datasets.
An auditing framework using justified representation from social choice theory is developed to evaluate question slates in expert panel deliberations, including efficient algorithms and comparisons of moderator, ILP, and LLM selections on historical data.
LLMs exhibit higher perplexity on far-right and nationalist party texts than social-democratic ones, consistent across models and languages with correlation to translation metrics.
Agora uses AI to ground policy discussions in real human voices and a small study shows it improves users' perspective-taking compared to numerical summaries alone.
The paper presents a conceptual design for an AI-augmented multiagent system that scaffolds structured sensemaking and argument visualization in educational and civic collaborative deliberation.
LLM facilitation in group charity allocation leaves consensus and participation equity unchanged while shifting specific allocations up to 5.5 points and increasing perceived trust.
A participatory design effort at FAccT used in-person sessions and Polis polling to co-create governance input and demonstrate scalable co-design for critical AI communities.
An AI system that elicits personal experiences and visualizes policy support increased perceived legitimacy and perspective-taking in collective decisions despite unfavorable outcomes.
citing papers explorer
-
Structure Before Collapse: Transient semantic geometry in next-token prediction
Semantic geometry emerges transiently early in next-token prediction training before collapsing to Neural Collapse symmetry in synthetic settings with latent semantic factors.
-
Efficient Elicitation of Collective Disagreements
Introduces the plurality matrix and a level-based stratification of disagreement measures, proving many require triple-wise information and providing two elicitation protocols that trade off participant numbers against cognitive load.
-
Embeddings for Preferences, Not Semantics
Synthetic training data designed to break the correlation between semantic and preferential signals in text embeddings provably improves preference prediction across 11 online deliberation datasets.
-
Question the Questions: Auditing Representation in Online Deliberative Processes
An auditing framework using justified representation from social choice theory is developed to evaluate question slates in expert panel deliberations, including efficient algorithms and comparisons of moderator, ILP, and LLM selections on historical data.
-
Large Language Models are Perplexed by some Political Parties
LLMs exhibit higher perplexity on far-right and nationalist party texts than social-democratic ones, consistent across models and languages with correlation to translation metrics.
-
Agora: Teaching the Skill of Consensus-Finding with AI Personas Grounded in Human Voice
Agora uses AI to ground policy discussions in real human voices and a small study shows it improves users' perspective-taking compared to numerical summaries alone.
-
Guided Sensemaking: Agents in Collaborative Deliberation
The paper presents a conceptual design for an AI-augmented multiagent system that scaffolds structured sensemaking and argument visualization in educational and civic collaborative deliberation.
-
Real-Time Group Dynamics with LLM Facilitation: Evidence from a Charity Allocation Task
LLM facilitation in group charity allocation leaves consensus and participation equity unchanged while shifting specific allocations up to 5.5 points and increasing perceived trust.
-
"Taking Stock at FAccT": Using Participatory Design to Co-Create a Vision for the Fairness, Accountability and Transparency Community
A participatory design effort at FAccT used in-person sessions and Polis polling to co-create governance input and demonstrate scalable co-design for critical AI communities.
-
AI and Collective Decisions: Strengthening Legitimacy and Losers' Consent
An AI system that elicits personal experiences and visualizes policy support increased perceived legitimacy and perspective-taking in collective decisions despite unfavorable outcomes.