Skip to content

Research separate from news

What could become important next

A preprint is a signal, not a finished product or an independently confirmed result. We therefore track research papers separately and show the actual state of supporting evidence for each one.

1381 published research events

Latest work

A significant claim from a single source is published only after further confirmation.

Research only one source so far

Study uses Fuzzy Cognitive Maps to generate synthetic health data

The study presents the generation of synthetic health data using Fuzzy Cognitive Maps running exclusively on CPU. On the Heart Disease dataset, the authors report an accuracy of up to 0.81 and AUROC of up to 0.90; the results are compared with the TVAE and Gaussian Copula methods.

arXiv cs.LG (Machine Learning) Original source ↗
Research only one source so far

Study compares evaluation methods for Romanian RAG systems using LLMs

The study adapted the Ragas framework for Romanian and created the AdminRo-Eval dataset. The authors report 96% agreement with human evaluation for Faithfulness when decomposing the evaluation with the Gemini 2.5 Pro model, and 90% for Answer Relevance when comparing answers.

arXiv cs.CL (Computation and Language / NLP) Original source ↗
Research only one source so far

Research survey classifies RAG methods along four axes

The study organizes RAG research according to retrieval efficiency, robustness and safety, interactive procedures, and multi-step reasoning. It compares methods, architectures, and evaluation approaches and describes persistent reliability and scaling challenges.

arXiv cs.AI (Artificial Intelligence) Original source ↗
Research only one source so far

CypherTurn study reveals low reliability of models in multi-step graph database querying

The study introduces CypherTurn with 721 conversations and 5 927 steps over 7 knowledge graphs. In evaluating 15 models, the best model achieved 64.7% query execution correctness; full-conversation correctness remained below 5%. A higher action budget did not resolve the gap in autonomous operation.

arXiv cs.CL (Computation and Language / NLP) Original source ↗
Research only one source so far

Study: probe-guided fine-tuning improves model safety and preserves their controllability

According to the authors, continuously updated probes during training reduce harmfulness and improve the truthfulness of models while preserving usefulness. The method shows a better safety-to-usefulness ratio than DPO and inference-time steering, and it preserves the ability to inspect internal representations.

arXiv cs.LG (Machine Learning) Original source ↗
Research only one source so far

Study presents Sieve and Sage for more reliable RALM system responses

The study presents Sieve and Sage: a lightweight module that filters out distracting source material before generating a response. Compared to single-stage approaches, the authors report an improvement in accuracy of up to 69.4 percentage points, Macro-F1 of 55.2 percentage points, and up to a 1.99x speedup.

arXiv cs.CL (Computation and Language / NLP) Original source ↗
Research only one source so far

Study presents FlexRouter for selecting mutually complementary LLM models

The study presents FlexRouter, which selects LLM models based on their capabilities and mutual complementarity. It determines the number of models dynamically. According to the authors, in RouterEval tests it achieves a higher probability of at least one correct answer and lower redundancy than the compared methods.

arXiv cs.LG (Machine Learning) Original source ↗