Semantic Alignment of AI Models: Concept Collapse, Checkpoint Dynamics, and Cross-Lingual Transfer

2026/08/03 by Tyler Ashoff, Jordan Rodu
Computer Science · #cs.CL #cs.LG

paper · pdf

Code available at github.com/tylerashoff/persiscope (PyPI: persiscope)

arxiv created 2026/08/03 · arxiv updated 2026/08/04

Abstract

Language model benchmarking is a difficult task. Outcome reasoning alone does not test the model's conceptualization of language and popular open-source benchmarks are quickly saturated or ingested as training data. It is important to test the model's output, but augmenting these tests by characterizing semantic structure gives more insight to how models relate abstract concepts. However, the high dimensional embedding spaces are not easy to interpret. This work demonstrates how topological methods can be used to rigorously compare these spaces to low dimensional and interpretable baselines like ontologies and curated knowledge graphs. These multi-modal alignment tests make it possible to track model adaptations and test phrase understanding across multiple languages.

Citations

• Persistent Convolution: A Topological Framework for AI Alignment Testing and Semantic Space Characterization
• Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models
• BIG-Bench Extra Hard
• Demonstrating specification gaming in reasoning models
• Alignment faking in large language models
• The Geometry of Categorical and Hierarchical Concepts in Large Language Models
• Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference
• Multilingual E5 Text Embeddings: A Technical Report
• M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation
• Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
• Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision
• The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets
• Language Models Represent Space and Time
• AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models
• Representation Engineering: A Top-Down Approach to AI Transparency
• Sparse Autoencoders Find Highly Interpretable Features in Language Models
• MADLAD-400: A Multilingual And Document-Level Large Audited Dataset
• Emergent Linear Representations in World Models of Self-Supervised Sequence Models
• Code of "Sirens' Whisper: Inaudible Near-Ultrasonic Jailbreaks of Speech-Driven LLMs"
• Jailbroken: How Does LLM Safety Training Fail?
• Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection
• Affective Coherence Monitoring for Transformer-Based Language Models
• Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task
• Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them
• Comparing multiple latent space embeddings using topological analysis
• No Language Left Behind: Scaling Human-Centered Machine Translation
• Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models
• Measuring Massive Multitask Language Understanding
• Language-agnostic BERT Sentence Embedding
• BLEURT: Learning Robust Metrics for Text Generation
• BART: Denoising Sequence-to-Sequence Pre-training for Natural Language\n Generation, Translation, and Comprehension
• Deep reinforcement learning from human preferences
• Stochastic Convergence of Persistence Landscapes and Silhouettes
• Efficient Estimation of Word Representations in Vector Space
• Representation Learning: A Review and New Perspectives
• Representation Learning: A Review and New Perspectives
• The control of the false discovery rate in multiple testing under dependency
• I.—COMPUTING MACHINERY AND INTELLIGENCE
• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding