AI-RADAR.IT · AI-RADAR.NET · AI-RADAR.TECH

News & analysis on local LLMs, stack & on-prem hardware.

🗺️ Sitemap

Complete navigation guide to all content on AI-Radar.it

📱 Core Pages

🌟 Comprehensive Guides

🛠️ Tools & Features

🏢 LLM On-Premise

🏗️ Architecture

🏭 Industry Scenarios

📂 Article Categories

📡 Feeds & API

📰 Recent Articles

HP Z4 G6i: Price Is Only the Surface of the Linux-Ready Signal

GLM-5.3: same base model, doubled cyber exploitation

HP Z4 G6i: Xeon 600, RTX and Linux-ready, but the top config tops $41,000

Linux kernel approaches 2,000 CVEs per release: the LLM factor

Micron: HBM Takes Three Times the Wafer Area of DDR5 at Equal Capacity

Linux 7.3 removes IBM and SGI drivers: the weight of LLM-generated fixes

Nvidia-Hugging Face: The Model Node Becomes a Hardware Asset

NeuronFuzz: fuzzing that looks inside LLM safety neurons

A Semantic Router for Labeled Property Graphs

PICasso and the Value of Physical Constraints: LLMs Alone Are Not Enough in...

EduRiskX: a neuro-symbolic route to early academic risk prediction

llama.cpp speeds up Qwen3.8-Flash-Next: 55 tokens/s on four RTX 3090s

Nvidia Moves to Acquire Hugging Face for $13 Billion

Nvidia, Hugging Face and the uncertain future of llama.cpp

Apodex 1.1: open agentic models land in multiple quantized formats

AMD jumps from ROCm 7.14 to ROCm 10.0 in an attempt to end versioning chaos

Apodex 1.1: open agentic models and quantization as a strategic signal

Qwen3.8-Flash-Next: Time to Update Those Benchmarks

GreenLeaf Law Embed Tiny and the Local Turn in Legal Retrieval

NVIDIA, Hugging Face and the Risk to Open Source

Different LLMs, Different Replies: Semantic Consistency Is Not Guaranteed

Decodable empathy directions don't guarantee reliable control in LLMs

GreenLeaf Law Embed Tiny: A Compact Embedding Model for Legal Retrieval

Tesla goes local in China, swapping Grok for ByteDance's Doubao LLM in smart cockpits

Nvidia in talks to acquire Hugging Face for over $13 billion

OpenAI and the Hugging Face case: more questions than answers

OpenAI: its agents learned to cheat before the Hugging Face hack

OpenAI releases official report on Hugging Face breach

GLM-5.3-Flash on Hugging Face: a name without a spec sheet

Qwen3.8-Flash-Next: A 125B Open-Weight Model for Memory-Constrained Accelerators

Goodfire Opens the Black Box: Making LLM Interpretability Accessible

IBM Granite 4.2: Three Self-Hosted LLMs and an Agentic Divide

AMD Adds a gfx1250-strict Target to Its LLVM Backend for Instinct MI450

Qwen3.8-27B Drops to 19.7 GB: What Changes for On-Prem LLM

Bigram entropy predicts weight growth in Transformer training

RENDER: Controlling Reader-Facing Evidence in LLM Memory Evaluation

Qwen3.8-27B in NVFP4: 19.7 GB and near-BF16 quality

China GPU race: Enflame IPO spotlights Tencent ties, losses and AI chip ambitions

Google Brings TPU 8t and 8i to Hot Chips 2026

LLVM 23.1: Zen 6 and Rigel support lands in the open-source compiler

MIRAVA raises $5 million and moves toward intelligent boat production

AI-generated bug reports become a denial-of-service attack on open-source maintainers

Linux 7.3: New Audio Hardware Support and the Weight of AI Patches

ToMoE and Dynamic Sparsity: Less Compute for Local LLMs

Dutch AI-native startup Neno raises €6.6M to expand agentic accounting across Europe

Khmer Semantic Search Still Rewards the Old BM25

KVBoost reuses chunk-level KV cache and cuts prefill without shared prefixes

Intel Crescent Island: 160GB to 480GB LPDDR5X for AI

Schools and AI: the traffic light disciplining chatbots without shutting them off

ToMoE Turns Dense LLMs into Mixture-of-Experts Without Fine-Tuning

View All Articles →