The LLM archive monitors model releases, quantization updates, reasoning capabilities, and real-world deployment implications for local and hybrid AI. We focus on what materially changes selection and operations: context windows, latency, memory footprint, licensing, and evaluation evidence across open and commercial families. This section is designed for teams that need dependable model intelligence, not hype cycles. Pair these updates with the LLM pillar and references to hardware constraints and framework integration.
Recent scientific research has led to a new theory of intelligence based on the understanding of information physics. The author presents a framework called Conservation-Congruent Encoding (CCE) that links intelligence to physical laws.
MetaJuLS is a meta-learning solution that introduces an universal approach to constraint propagation for structured inference in large language models. This approach achieves 1.5-2.0x speedups over GPU-optimized baselines while maintaining within 0.2% accuracy of state-of-the-art parsers.
Un nuovo approccio per integrare la ricerca e la ragione negli LLMs. Il metodo introduce una strategia di recupero del sapere che si concentra sulla struttura logica delle conversazioni, migliorando così il rendimento dei modelli.
Researchers have proposed a new method for adapting language models to specific tools, called RIMRULE. This approach uses dynamic rule injection and an MDL objective to favor generality and conciseness of rules.
The decline of AI system performance is a fundamental problem to improve efficiency and reliability. Temporal knowledge graph reasoning is crucial for future AI applications.
The Pat-DEVAL team has presented a new evaluation framework for patent descriptions, called Chain-of-Legal-Thought Evaluation. This approach uses large language models to evaluate the structural coherence and statutory compliance of patent descriptions.
A research group has introduced a new form of optimal identification with limited error control, overcoming the limitations of existing methodologies.
A new attack discovered on the LLM composition system may compromise model security.
La gestione del carico di lavoro operativo delle risorse umane è critica nel sistema di consegna last mile. L'approccio multi-algoritmo proposto utilizza una combinazione di considerazioni di distanza e carico di lavoro per ottimizzare l'allocazione delle consegne ai lavoratori.
A team of scientists has discovered an ancient cremation in Africa, dating back over 9,500 years. The event was found at a site in Malawi and represents one of the oldest examples of cremated remains found in a pyre.
Disinformation Floods Social Media After Nicolás Maduro's Capture
The New York Times reports that US intelligence agencies invaded Venezuela and captured President Nicolás Maduro. But ChatGPT, the Meta LLM model, disputes this account.
La società di AI OpenAI sta ristrutturando alcuni team per sviluppare prodotti hardware basati su tecnologie audio, con l'obiettivo di migliorare la precisione e la velocità dei modelli. La nuova piattaforma sarà focalizzata sull'audio e si spera che potrà spingere gli utenti a utilizzare più frequentemente l'interfaccia vocale.
Scientists Identify Remains of the Earliest Human Ancestor
India orders Musk’s X to fix Grok over “obscene” AI content
Mercor, a three-year-old startup, has become a $10 billion middleman in AI's data gold rush. The company connects AI labs like OpenAI and Anthropic with former employees of Goldman Sachs, McKinsey, and white-shoe law firms, paying them up to $200 an hour to share their industry expertise and train the AI models that could eventually automate their former employers out of business.
Micron secures $318 million Taiwanese subsidy for HBM R&D as AI memory arms race intensifies — three-year project aims to develop leading-edge, high-performance memory
Nvidia ha investito oltre 100 startup nel campo dell'intelligenza artificiale in due anni. Ecco i suoi maggiori investimenti.
A dispute has flared up inside the Debian project after a senior maintainer criticized the distribution’s bug tracking system as outdated and increasingly unworkable for modern software development
In 2026, here's what you can expect from the AI industry: new architectures, smaller models, world models, reliable agents, physical AI, and products designed for real-world use.