Sparse Autoencoders
PulseAugur coverage of Sparse Autoencoders — every cluster mentioning Sparse Autoencoders across labs, papers, and developer communities, ranked by signal.
- instance of Gotit.pub 90%
- instance of Saessolsheim 90%
- used by Vision Transformers 80%
- used by Gotit.pub 70%
- developed by Gotit.pub 70%
- used by Saessolsheim 70%
- used by UniPro 70%
- instance of SAE International 70%
- developed Vision Transformers 70%
- developed by Vision Transformers 70%
- used by GPT-2 small 70%
- used by InternVL3.5 70%
- 2026-05-25 research_milestone Researchers published a paper detailing a new method for multilingual language steering in LLMs using sparse autoencoders. source
- 2026-05-21 research_milestone Researchers published a paper detailing a new method for multilingual steering in LLMs using sparse autoencoders. source
16 day(s) with sentiment data
-
New framework decodes user embeddings into interpretable behavioral modes
Researchers have proposed a new framework to interpret dense and opaque user embeddings, which are crucial for recommendation and personalization systems. The proposed method suggests that these embeddings are superposi…
-
SteerGR framework enables controllable game recommendations using sparse autoencoders
Researchers have developed SteerGR, a new framework designed to provide explicit control over generative recommendation systems, particularly for game discovery. This framework utilizes sparse autoencoders to link human…
-
Sparse Autoencoders Reveal Fine-Grained Specialization in MoE Experts
Researchers have developed a new method using Sparse Autoencoders (SAEs) to analyze the internal workings of Mixture-of-Experts (MoE) models, specifically Qwen3-30B-A3B. This approach reveals that individual experts wit…
-
New framework clarifies sparse visual features in neural networks
Researchers have introduced Reconstructive SAE Intervention (RecSAEI), a novel framework designed to clarify the semantic meanings of sparse visual features within neural networks. Traditional methods of interpreting th…
-
TraceRouter++ framework realigns LLM safety pathways using optimal transport
Researchers have introduced TraceRouter++, a novel framework designed to enhance the safety of large foundation models (LFMs) by realigning causal semantic routing pathways. Unlike existing methods that focus on localiz…
-
New "brain scan" technique reveals LLM reasoning flaws
Researchers have developed a method using sparse autoencoders (SAEs) to analyze the internal workings of large language models, akin to a "brain scan." This technique allows for the interpretation and targeted manipulat…
-
LLM personas alter internal representations, impacting safety evaluations
A new research paper explores how large language models (LLMs) reorganize their internal representations when prompted to adopt specific political personas. The study found that conditioning models on liberal or conserv…
-
New method analyzes sparse autoencoder behavior with attribution graphs
Researchers have developed a new method using Sparse Autoencoder Attribution Graphs (SAEAGs) to analyze the behavior of sparse autoencoders (SAEs). These graphs help predict the impact of feature ablations and are more …
-
New framework enables comparison of sparse autoencoders across language models
Researchers have developed a new framework to compare sparse autoencoders (SAEs) by connecting feature correspondence, dictionary differences, and uncertainty from evaluation data. This framework utilizes an energy-weig…
-
Sparse Autoencoders Enhance Histopathology Model Interpretation
Researchers have developed sparse autoencoders (SAEs) to better interpret the embeddings generated by histopathology foundation models. These SAEs decompose complex embeddings into simpler, sparse latent features that c…
-
Mechanistic interpretability applied to end-to-end self-driving models
Researchers have applied mechanistic interpretability techniques to an end-to-end autonomous driving model, utilizing Sparse Autoencoders (SAEs) to analyze its internal activations. This approach revealed that SAE laten…
-
New research probes causal influence of AI model features on errors
Researchers have investigated the causal influence of Sparse Autoencoder (SAE) features on model errors in multiple-choice question answering tasks. Using Gemma 3 1B and 4B models, they developed a pipeline to filter ou…
-
LLM reasoning skills questioned in new Tic-Tac-Toe study
A new research paper explores whether large language models (LLMs) trained with reinforcement learning (RL) truly develop reasoning abilities or merely exploit patterns in prompts. By analyzing LLMs trained on Tic-Tac-T…
-
Interpreting ESMFold's protein folding mechanisms with sparse autoencoders
Researchers have developed a method using sparse autoencoders to interpret the internal workings of ESMFold, a neural network used for protein structure prediction. By analyzing the model's representations, they identif…
-
New framework adapts interpretable features for generative recommendation
Researchers have developed a new framework called API that adapts pretrained interpretable features for generative recommendation systems. This approach utilizes pretrained sparse autoencoders to expose human-interpreta…
-
Sparse Autoencoders enhance interpretability in medical imaging foundation models
Researchers have developed Sparse Autoencoders (SAEs) to make medical imaging foundation models more interpretable for clinicians. By training SAEs on embeddings from BiomedParse and DINOv3 using a large dataset of CT a…
-
New TaxoSAE method enhances LLM interpretability with taxonomic logic
Researchers have introduced TaxoSAE, a novel approach to Sparse Autoencoders (SAEs) that incorporates taxonomic logic to improve the interpretability and control of knowledge encoded within large language models (LLMs).…
-
New TAME framework combats VLM reasoning obfuscation
Researchers have developed a new framework called TAME to address chain-of-thought (CoT) obfuscation in vision-language models (VLMs). This obfuscation occurs when models achieve high accuracy but produce reasoning trac…
-
New TopKLoRA method enables interpretable AI model fine-tuning and backdoor removal
Researchers have developed TopKLoRA, an interpretable fine-tuning method designed to identify and remove malicious backdoors in AI models. This technique, an extension of prior work, activates only the largest adapter l…
-
New method precisely erases concepts from diffusion models
Researchers have developed a new method called Concept Acupuncturist (ConceptACP) to efficiently and precisely remove specific concepts from diffusion models. This approach addresses the challenge of concept entanglemen…