steering vectors
PulseAugur coverage of steering vectors — every cluster mentioning steering vectors across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
Activation steering improves ASR performance and fairness
Researchers have explored activation steering as a method to enhance both performance and fairness in Automatic Speech Recognition (ASR) systems. Their investigation focused on two techniques: paired activation subtract…
-
New 'Semantic Overlays' Technique Enhances LLM Security Against Prompt Injection
Researchers have developed a new technique called Semantic Overlays to combat prompt injection attacks in language models. This method introduces a non-textual channel to the model's input, allowing it to understand the…
-
New method reduces safety risks of LLM steering vectors
Researchers have developed a method to mitigate the safety risks associated with steering vectors in large language models. These vectors, used for controlling model behavior, can unintentionally degrade safety mechanis…
-
New research reveals "inverted" steering vectors in LLMs
Researchers have identified an "inverted detection-control phenomenon" in steering vectors (SVs), a technique used to influence the output of large language models. This phenomenon occurs when highly discriminative SVs,…
-
AI alignment research tackles reward hacking with new techniques
Researchers are exploring methods to prevent AI models from exploiting reward functions, a phenomenon known as reward hacking. One approach involves using steering vectors to guide gradient routing, aiming to isolate un…
-
Soft prompt distillation enhances on-device LLM safety
Researchers have developed a new method for making large language models safer and more efficient for use on devices with limited resources. The technique involves using "soft prompts" combined with distillation to tran…
-
Steering vectors offer direct control over LLM tone, bypassing prompt limitations
Prompt engineering is often ineffective for controlling the tone of large language models because behavioral traits are encoded in the model's internal state, not just its input prompts. A technique called activation st…