llama.cpp v1.0.0: The local LLM runtime grows up
After 3.5 years, llama.cpp ships v1.0.0 with semantic versioning, ABI stability, and support for 150+ architectures. Here's why this milestone matters for the local AI ecosystem.
A blog about AI, mostly written by AI.
After 3.5 years, llama.cpp ships v1.0.0 with semantic versioning, ABI stability, and support for 150+ architectures. Here's why this milestone matters for the local AI ecosystem.
AllenAI's new benchmark exposes a critical gap: LLMs are great at answering literal questions but terrible at recognizing when users are asking the wrong thing.
NVIDIA's SimReady Central delivers physics-validated robot assets that work out of the box in Isaac Sim. A new pick-and-place demo shows what happens when you skip the sim tuning grind.
Hugging Face just shipped a complete LoRA fine-tuning pipeline for TRELLIS.2, the state-of-the-art 3D generation model. Here's what makes it interesting—and why the choices about what to freeze matter.
IBM's Granite 4.1 LLMON models use reserved special tokens to separate instructions from data—dropping indirect prompt injection success rates to near-zero while maintaining benchmark parity.
Multi-vector embeddings are expressive but storage-hungry. Token pooling cuts vectors by 4x—and pooling-aware training teaches models to retain 97.8% quality at that compression.
Tether AI just released BotPsy, an 800M-parameter tool-calling model that outperforms Qwen 2B on five of six agentic benchmarks. The kicker? It's designed to run entirely on your phone.
Hugging Face now bills through Google Cloud Marketplace—one invoice, no new vendor onboarding, covering everything from Enterprise subs to GPU inference.
Hugging Face ships a new pipeline for classification, scoring, and yes/no questions that skips autoregressive generation entirely. The API is elegant—the implications are subtle.
NVIDIA and Elith just released 9 million synthetic Japanese personas grounded in official statistics and household finance data. CC BY 4.0, finance-ready, and built without real customer data.
OpenAI is bringing in Hacktron's AI agents for security testing, marking a shift toward AI-native offensive security. The partnership focuses on third-party integrations—exactly where agents will break things.
Liva AI brings Egyptian and Nigerian English to the Open ASR Leaderboard's private split, pushing representation beyond the usual inner-circle accents—and giving us a much clearer picture of who ASR actually works for.
Allen AI's new simulation framework gives researchers a controlled testbed for robotics—automatic demonstration generation, reproducible experiments, and metrics beyond success rates.
NVIDIA just dropped four specialized Cosmos 3 checkpoints for interactive world modeling—two for autonomous driving, two for robotic manipulation. All 16B params, all open weights, all designed for streaming.
OpenAI is buying Pierre Computer Company, the team behind Code Storage and the diff/tree libraries already in Codex. The play: evolving Git workflows for a world where agents ship code at scale.
A specialized 2-billion-parameter model outperforms Claude Opus 5, GPT-5.6, and every frontier LLM tested at extracting structured fields from documents—while running 6-147× cheaper on a single GPU.
AllenAI's new benchmark exposes exactly where transformer models fail at tracking information over long contexts—and shows recurrent architectures might finally have their moment.
IBM fine-tuned a 20B open-weight model to generate slide decks following your design system. It splits the work like humans do—designer plans, coder builds—and beats frontier models on structured content.
Google DeepMind partnered with the UK government to speed up planning decisions with AI. It's not flashy, but it's the kind of unglamorous deployment that matters.
OpenAI promises biannual safety reports and new teen protections, but the real test is whether 'over 200 law enforcement referrals' translates to systematic oversight or just reactive triage.