Linear Semantic Control
No mentions found
This entity hasn't been tracked yet, or Iris is still building its knowledge base.
Related Articles from SNS
LiSeCo: Linear Semantic Control for Language Generation
arXiv:2405.15454v4 Announce Type: replace Abstract: The prevalence of Large Language Models (LLMs) in critical applications highlights the need for controlled language generation methods that are both computationally efficient and enjoy performance guarantees. To address this need, we use a common model of concept semantics as linearly represented in an LLM's latent space. In particular, we take the view that natural language generation traces a trajectory in this continuous semantic space,...
Genie 4D: Semantic-Prior-Guided 4D Dynamic Scene Reconstruction
arXiv:2604.09877v2 Announce Type: replace Abstract: At the intersection of computer vision and robotic perception, 4D reconstruction of dynamic scenes connects low-level geometric sensing with high-level semantic understanding. We present Genie 4D, a framework that turns hand-held phone capture into a semantically grounded, action-controllable 4D world model. Genie 4D couples a real-time visual-inertial Gaussian splatting front-end for metric geometry with a feed-forward 4D backbone...
The Information Geometry of Softmax: Probing and Steering
arXiv:2602.15293v2 Announce Type: replace Abstract: This paper concerns the question of how AI systems encode semantic structure into the geometric structure of their representation spaces. The motivating observation is that the natural geometry of these representation spaces should reflect the way models use representations to produce behavior. We focus on the important special case of representations that define softmax distributions.
Rethinking Search as Code Generation
Rethinking Search as Code Generation Evolving search from monolithic services to programmable primitives for the era of agent harnesses. Search is a core primitive for AI systems. Frontier models grow more capable by the month, but they still need access to fresh, accurate, and well-curated knowledge from the wider world.
Subspace-Decomposed JEPAs: Disentangling Progression and Content in Latent World Models
arXiv:2605.31111v1 Announce Type: new Abstract: Joint-Embedding Predictive Architectures (JEPAs) learn compact latent world models by predicting future embeddings, but no single coordinate of the latent is designated to encode task progression. We carve the JEPA latent into two orthogonal subspaces with disjoint roles: a low-dimensional progression subspace shaped by a cosine-margin triplet loss, and a high-dimensional content subspace regularised by the existing SIGReg objective of LeWM. We...
Unfolding Generative Flows with Koopman Operators: Trajectory-Preserving Linearization
Announce Type: replace Abstract: Continuous Normalizing Flows (CNFs) enable elegant generative modeling but remain bottlenecked by their iterative nature requiring costly sampling and lacking interpretability of the intermediate states. Recent approaches accelerate sampling by straightening trajectories or distilling endpoints, yet they treat the original generative process as a black box, discarding the teacher's intermediate dynamics. We propose a fundamentally different perspective:...
AlgoTouch: An Execution-Centered Approach to Incremental Construction of Imperative Programs
arXiv:2606.03349v1 Announce Type: new Abstract: Program construction in imperative languages remains largely based on writing textual code that specifies sequences of instructions operating on program data. This approach requires developers to anticipate the effects of instructions on evolving data states, which increases cognitive load and the likelihood of errors during early and incremental development. This paper presents AlgoTouch, an execution-based system for incremental construction...
Ablation-Reversible Heads Don't Transfer: A Stress Test for Mechanistic Role Claims in Transformers
Announce Type: new Abstract: In mechanistic interpretability, attention heads are commonly elevated to role claims (e.g., "this head represents addition") when they are necessary for a behavior, encode it linearly, and recover that behavior when restored after ablation. We show this evidence is insufficient: across three 7-8B instruction-tuned models and five computation families, heads passing all three checks routinely fail to transfer the computation when their activations are patched...
Step-Level Sparse Autoencoder for Reasoning Process Interpretation
arXiv:2603.03031v2 Announce Type: replace Abstract: Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-Thought (CoT) reasoning. However, their reasoning patterns remain too complicated to analyze. While Sparse Autoencoders (SAEs) have emerged as a powerful tool for interpretability, existing approaches predominantly operate at the token level, creating a granularity mismatch when capturing more critical step-level information, such as reasoning...
ECI: Effective Contrastive Information to Evaluate Hard-Negatives
arXiv:2603.20990v2 Announce Type: replace Abstract: Hard-negative source selection for dense retrieval is usually decided only after fine-tuning and downstream evaluation. We propose Effective Contrastive Information (ECI), a training-free diagnostic that ranks candidate negative sources using frozen target-encoder embeddings. ECI is training-free, not label-free: each scored example requires a query, a labeled positive, and an explicit candidate negative.