New studies test LLMs on long-context summarization, retrieval with BM25, digital twins, epidemic forecasting, neural decoding, scientific software recognition and self-directed research.
A study on 3,000 Khmer documents shows BM25 outperforms dense retrieval and Qwen2.5-assisted query expansion. The larger model produces better expansions but still introduces noise. A signal for teams building low-cost local search.
Research shows that organizing personal information into structured schemas improves predictive accuracy for LLM-based digital twins. On homogeneous benchmarks the gain is +1.91 percentage points, but fixed schemas fail on heterogeneous tasks. An aut...
A study on 190 weekly Ontario case counts shows ARIMA reacts quickly to turning points but degrades over longer horizons, while random forest and XGBoost offer more stability. The MLAMA ensemble, with performance-based weights varying by horizon, ach...
A 450M parameter VLM goes from 1 to 44 out of 100 after fine-tuning on 50,000 browser screenshots. The source does not specify the metric or infrastructure, but the jump shows compact models can become usable on domain-specific visual data. For those...
A hybrid framework combines lexical signals and SciBERT semantics to identify software and database names in biomedical literature. Trained with a pipeline mixing citation extraction and LLM-assisted distillation, it outperforms specialized methods a...
Microsoft Research has released Skala 1.1, a deep-learning exchange-correlation functional. Trained on 2.5 times more data, it lowers the weighted average error to 2.8 kcal/mol on GMTKN55 while retaining meta-GGA cost. Native integration in CP2K, wit...
LongNovel is a multi-scale, bilingual Chinese-English benchmark for detecting hallucinations in long novel summaries. Built on 29 Chinese novels from 16k to 100k tokens and BookSum data, it identifies eight hallucination types. The test set was manua...
A new framework aligns brain and text embeddings in a shared semantic space for retrieval-based decoding, separating neural signal from LLM reconstruction. MD-SigLIP uses duplicate-aware contrastive learning and a listwise margin term to enforce rank...
A Princeton-led study put Claude Opus 4.8 agents to work on unpublished NeurIPS 2026 research questions. The agents handled engineering tasks but lacked the judgment and creativity needed for open-ended research, and both papers were rejected. The fi...
