Researchers are developing new methods to optimize Large Language Model (LLM) routing, aiming to balance inference costs with user satisfaction and Service Level Agreements (SLAs). SLARouter, an online algorithm, learns from sparse user feedback to achieve cost optimality and SLA compliance, reducing costs by up to 2.2x. Separately, RouteJudge offers an open platform for evaluating LLM routing systems, focusing on router-level decision quality and providing a toolbox for developing and comparing routing algorithms. Additionally, a new attack, the Forced Deferral Attack (FDA), has been identified that manipulates multimodal LLM cascades by forcing queries to more expensive models, highlighting a new security vulnerability. AI
IMPACT New routing strategies and security analyses could significantly impact LLM operational efficiency and robustness.
RANK_REASON The cluster contains multiple research papers detailing new algorithms and frameworks for LLM routing and security.
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Forced Deferral Attack
- Gotit.pub
- Hugging Face
- MLLM cascades
- Multimodal Large Language Models and Tunings: Vision, Language, Sensors, Audio, and Beyond
- ScienceCast
- ORBIT
- RouteJudge
- Herbert Woisetschläger
- IArxiv
- Large Language Models
- LLM cascades
- SLARouter
AI-generated summary · Google Gemini · from 5 sources. How we write summaries →
