Sci-Rho: New Multilingual Benchmark Tests AI Model Robustness on STEM Problems
Researchers have introduced Sci-Rho, a new dynamic benchmark comprising 4,242 expert-crafted problem templates across five STEM subjects and seven languages, designed to test the robustness of vision-language models (VLMs) under varied problem conditions. Unlike existing symbolic benchmarks, Sci-Rho incorporates visual grounding and multilingual coverage, generating 42,420 total problem instances by systematically varying numerical values, visual patterns, and other parameters. The benchmark exposes a meaningful gap between models' average accuracy and their worst-case accuracy, raising concerns about how reliably current AI systems generalize across problem variations.
Sci-Rho (Science Rhobustness) is a newly proposed multilingual, visually-grounded symbolic benchmark aimed at more rigorously evaluating the robustness of state-of-the-art vision-language models on STEM reasoning tasks. The benchmark spans five subjects and seven languages, with 4,242 problem templates — 606 per language — crafted by domain experts including Olympiad medalists, each implemented as executable Python code that generates diverse but equivalent problem instances. In total, 42,420 instances are produced by varying numerical values, visual patterns, geometric shapes, color schemes, and function types, with each instance paired with reasoning steps and ground-truth solutions. Evaluations of 17 leading VLMs revealed a notable discrepancy between average accuracy and worst-case accuracy — the proportion of templates a model answers correctly across every generated variation — suggesting models may perform well on average while failing systematically on specific variants. Smaller models showed pronounced performance degradation across languages, while larger proprietary models demonstrated greater robustness. Step-level evaluation using F1 scores confirmed the same trend, and analysis of attention heads in one VLM revealed substantial cross-lingual variation in how much attention is allocated to image tokens versus text tokens. The authors argue their work underscores the need to move beyond static benchmarks when assessing VLM quality.
What's missing
It is unclear how the seven languages were selected, whether they represent balanced typological diversity, and whether the domain expert pool introduces any systematic subject-matter or cultural biases in problem design. The gap between worst-case and average accuracy is reported but the practical threshold at which such a gap becomes a meaningful deployment concern is not discussed.
What different sources said
- arXiv cs.CLCenter
KCSAT-ML: Probing Reasoning Models with Nationwide-Cohort Human Difficulty
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.