The study explores the effectiveness of representation engineering versus behavioral alignment methods in LLM safety, finding that while behavioral methods like DPO offer stronger control, representation steering can be competitive in low-data scenarios. Specialized text monitors outperform representation probes in detection accuracy, but representation-based approaches offer cost advantages.
The chart will appear after repeat observations. The current metric comes from the source.
2 source votesReal observations only. History before source connection is not reconstructed.
This research provides insights into when representation engineering can complement or replace behavioral alignment methods for AI safety, helping to inform the design of more robust safety mechanisms.
This paper introduces FuseReg, a method that replaces heuristic layer fusion in representation autoencoders (RAEs) with training over random subsets of encoder layers. The approach reduces the reconstruction-generation gap by improving robustness to layer fusion choices, achieving higher PSNR and lower generation FID scores without modifying the pretrained encoder.
4 days agoDaily PapersThe paper explores the linearity in Large Language Models (LLMs) by showing that combining inputs from different text streams leads to a superposition of next-token distributions. It suggests that this linearity is an inherent property of the Transformer architecture and can be restored through fine-tuning, allowing for generating two coherent continuations from one forward pass.
5 days agoDaily PapersThis paper introduces a method to enhance large language models' (LLMs) ability to learn from context by using perturbed public documents. The approach involves rewriting documents, generating reasoning-based questions, and training a student model on these synthetic samples, leading to improved performance on CL-bench and broader transfer to tasks like long-context understanding and reasoning.
2 days agoDaily PapersThe paper introduces Skill2Env, a framework that generates executable environments from skills by using agent capability demands to create task blueprints with objectives, challenges, and evaluation criteria. It also proposes Iterative Task Hardening to improve task difficulty based on solver execution evidence, leading to improved agent performance across benchmarks.
2 days agoDaily PapersThe paper introduces PISA, a block-sparse attention mechanism with log-linear complexity, designed to scale language models for long contexts. It uses a pyramid Top-K selection strategy to efficiently identify relevant keys, reducing computational cost compared to traditional methods.
4 days agoDaily PapersThis paper reviews memory mechanisms in autoregressive (AR) video generation, focusing on how historical information influences future generations despite limited context windows. It organizes the literature through five perspectives: forms, functions, operations, learning, and evaluation, highlighting challenges like resource-aware memory architectures and standardized evaluation.
6 days ago