The Who's Who of AI

“Some new data from Anthropic looking at the impact of AI on the sciences by field: "With current capabilities, if adopted widely, AI could take on 27–52% of the researcher time needed to produce a new study, with a median of 38% across fields..." Here: www.…” evidence ↗

“Researchers offer a new view on diffusion transformer depth profiles, revealing how training setups and prediction targets shape model behavior—enhancing applications like REPA for improved generative performance in image generation. https://arxiv.org/abs/2…” evidence ↗

“A study shows that in multi-agent LLM systems, information disclosure doesn't always improve collective outcomes; finer details can lower efficiency, highlighting the need to optimize both model selection and information policies to maximize gains. https://…” evidence ↗

“Jaival Chauhan, Vikram Singh, Pruthwik Mishra Automated Video-to-QA Pipeline: End-to-End Question Generation and Verification Using Large Language Models https://arxiv.org/abs/2611.07755” evidence ↗

“Ayal Klein Decodability Is Not Validity: Evaluating Nuisance Sensitivity in Language Model Concept Probes https://arxiv.org/abs/2611.07805” evidence ↗

“Edoardo Vaira, Georgios Chavales, Pierre Nugues Different Corpora, Different Skills? A Paired Study of Small Language Models https://arxiv.org/abs/2611.07812” evidence ↗

“I'm working on expanding the perspective that we wrote about in the ElicitationGPT paper (see the section on "Algorithmic Artificial Intelligence") that relates developing reliable AI systems to decades old work on Human Computation. pubchat.org/chat/arxiv/…” evidence ↗

“ProMoral-Bench introduces a benchmark for assessing prompts in large models' moral judgments and jailbreak resistance, revealing compact, exemplar-guided prompts outperform complex ones. This study sets a standard for ethical AI assessments, enhancing model…” evidence ↗

“Muon, a new training method, shifts classic stability concepts in gradient descent, emphasizing the independence of loss balance and temporal alignment. It improves training dynamics while preserving coherence, marking a significant advancement. https://arx…” evidence ↗

“Researchers unveiled LowRAM, a KV cache compression method achieving 26.9× compression and 94.5% baseline performance, enhancing autoregressive decoding efficiency. https://arxiv.org/abs/2611.07025” evidence ↗

“Hongyang Su, Beibei Kong, Lei Cheng, Chengxiang Zhuo, Zang Li Do Recommenders Need to Think in Words? Latent Reasoning for Efficient LLM-Based Recommendation https://arxiv.org/abs/2611.07866” evidence ↗

“Research emphasizes a shift in data selection for fine-tuning language models, favoring coverage over difficulty. A novel clustering method surpasses traditional methods, achieving better performance on varied datasets. https://arxiv.org/abs/2611.07323” evidence ↗

“SA VeR2 is a method for safety alignment in Large Reasoning Models, using verifiable rewards and gradient projection to increase safety while retaining reasoning skills. Findings show enhanced safety across models with up to 671B parameters. https://arxiv.o…” evidence ↗

“A study shows supervised fine-tuning boosts model performance, while verifier-based reinforcement learning yields modest gains, especially during distribution shifts. This challenges beliefs about RL's effectiveness in AI training, emphasizing budget alloca…” evidence ↗

“Md. Rafiul Biswas, George Mikros, Wajdi Zaghouani Compression Through a Context Lens: Evaluating Adaptation, Quantization, and Distillation in Arabic Legal LLMs https://arxiv.org/abs/2611.07988” evidence ↗

“Jiacheng Cai, Tao Zhang, Siqi Liang, Yun Dai, Ziqian Zeng Beyond K/V Caches: Cross-Model State Handoff in Hybrid Linear-Attention LLMs https://arxiv.org/abs/2611.08133” evidence ↗

“Jaewon Cho, Minseo Yoon, Jinyoung Park, Jinwoo Seo, Jaewon Chu, Alice Oh, Hyunwoo J. Kim Knowing When to Teach: Learning from Teachable Prefixes in Multilingual Reasoning Distillation https://arxiv.org/abs/2611.08179” evidence ↗

“Russell Yang, Ruishi Chen, Pierce Kelaita, Riya Ranjan, Sibo Ma, Charles Dickens, Matthew Guillod, Megan Ma, Julian Nyarko JudgmentBench: Comparing Rubric and Preference Evaluation for Quality Assessment https://arxiv.org/abs/2605.25240” evidence ↗