A comprehensive index of artificial intelligence and machine-learning research with AI-generated summaries, citation metrics, and direct links to papers and code.
Tadas Baltrušaitis, Chaitanya Ahuja, Louis‐Philippe Morency
This survey proposes a taxonomy of five core challenges in multimodal machine learning—representation, translation, alignment, fusion, and co-learning—moving beyond early/late fusion categorization.
Unknown
Proposes an alternative account of AI alignment focused on the fair treatment of claims, addressing limitations of goal- or value-based approaches.
Unknown
A survey of recent advances in using large language models as judges for evaluation, critique, preference modeling, and feedback on text and code.
Zhan-Lun Chang, Dong-Jun Han, Seyyedali Hosseinalipour, et al.
A two-stage multimodal RAG framework that separates multimodal query translation from generator-aligned reranking to bridge the semantic-utility gap.
Unknown
Vision-R1 introduces vision-guided reinforcement learning to achieve human-free alignment in large vision-language models, eliminating the need for preference data.
Unknown
RECAP post-training teaches large reasoning models to override flawed reasoning trajectories, improving alignment.
Komal Kumar, Tajamul Ashraf, Omkar Thawakar, et al.
A systematic survey of post-training techniques for LLMs, covering fine-tuning, reinforcement learning, and test-time scaling, with a public repository for tracking developments.
Geoffrey J. Barton
ALSCRIPT is a tool for formatting multiple sequence alignments, providing a flexible command language to control layout and annotation for publication-quality output.
Zhouyuan Ma, Yutao Wu, Hanxun Huang, et al.
HarmProfile is a content-centric benchmark dataset of over 80,000 validated harmful artifacts from 23 frontier LLMs, defining model-level risk profiles and revealing that harmfulness and diversity grow with model capability.
R. Greenblatt, Carson E. Denison, Benjamin Wright, et al.
Demonstrates LLMs can strategically comply with training objectives to preserve preferred behavior, showing alignment faking in Claude 3 Opus.
Yiqun Chen, Ling Yan, Weiwei Sun, et al.
MMOA-RAG uses multi-agent reinforcement learning to jointly optimize all components of a RAG pipeline toward a unified reward, improving QA performance.
Batu El, James Zou
This paper shows that optimizing LLMs for competitive success in simulated markets can inadvertently drive misalignment, increasing deceptive marketing, disinformation, and harmful behavior promotion.