📁 LLM

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.

CodeGEMM is a new approach to optimize performance of large models (LLMs) using quantization. This work presents a new GEMM kernel that replaces dequantization with precomputed inner products between centroids and activations stored in a lightweight codebook.

2025-12-23 Fonte

I grandi modelli di linguaggio (LLM) hanno reso possibile l'utilizzo di sistemi multi-agenti (MAS) in cui molti agenti discutono, criticano e coordinano per risolvere compiti complessi. Tuttavia, la maggior parte degli LLM-based MAS adotta grafici pieni o reti sparse, con poca guida strutturale. Questo articolo esplora come le reti di piccolo mondo possano essere utilizzate per stabilizzare i sistemi multi-agenti.

2025-12-23 Fonte

A new publication analyzes the faithfulness and stability of neuron explanations to ensure trustworthy interpretation. The proposed method offers a clear direction for future research in this critical field.

2025-12-23 Fonte

The latest version of ChatGPT's Llama model introduces a new feature that allows users to directly influence the enthusiasm level of their conversations. This innovation enables more personalized interactions with the platform.

2025-12-23 Fonte
📁 LLM AI generated

I rischi nascosti degli LLM

Gli LLM stanno rivoluzionando l'industria tecnologica, ma anche con loro vengono associate nuove sfide di sicurezza. Un recente rapporto dell'OWASP elenca i rischi più critici da prioritare.

2025-12-23 Fonte