Make sense of what AI is becoming.

Xorricane solves XOR‑heavy SAT instances with XNF and XLIN

The European Commission will require providers of general‑purpose AI models to notify regulators by Aug 27 2026.

A new systematic review catalogs 257 text‑generation studies covering five core tasks and highlights persistent challenges such as bias, hallucinations, and evaluation subjectivity.

Researchers propose a way to lower‑bound classifier unfairness without accessing individual predictions or model internals.

Researchers unveil Xorricane, a SAT solver using XOR‑OR‑AND normal forms that improves efficiency on cryptographic problems.

Researchers propose a way to estimate the worst-case bias of black-box classifiers without accessing individual predictions.

The JAIR review covers 257 papers on five core text‑generation tasks and identifies nine recurring challenges including bias and hallucinations.

JAIR released three papers covering fairness auditing with population statistics, XOR-based SAT solving, and neuro‑symbolic argument quality assessment.

A recent JAIR paper warns that neuro‑symbolic systems can misground concepts, undermining trust.

Les chercheurs montrent que répondre à des requêtes régulières à deux sens sous règles linéaires ou gardées a une complexité en données NL‑complète mais peut atteindre 2ExpTime en scénario combiné, ce qui influence les applications de graphe de connaissances biomédicales.

Researchers introduce DyMAB, a dynamic multi‑armed bandit algorithm that enhances ALNS-based MAPF solvers by adapting neighborhood selection over time.

Researchers show linear existential rules keep path query answering in NL data complexity while guarded rules push it to 2ExpTime combined complexity.