Llama is a topic tracked in our intelligence system with 5 linked articles.
Five major publishers and an author sue Meta over allegedly training Llama on copyrighted works without permission, citing copying from pirate sites.
A practitioner-driven, unverified set of forum anecdotes on OCR/IDP pipelines highlights production gaps, the rise of hybrid architectures, table extraction challenges, and regulatory/data-residency constraints shaping enterprise deployment.
A simple self-distillation method yields substantial code-generation gains across LLMs, boosting pass@1 on LiveCodeBench v6 from 42.4% to 55.3% and generalizing across Qwen and Llama models at 4B/8B/30B scales.
SwiftLM is a native Apple Silicon LLM server (OpenAI-compatible) with TurboQuant KV cache compression and SSD expert streaming, plus iOS/macOS platform-aware management and a licensing migration to MIT.
Arcee AI, a 30-person US startup, released Trinity, a 400-billion-parameter open-source LLM, claiming it’s among the largest such models from a US company.
ADL study finds Grok worst at countering antisemitic content across six LLMs, with Claude best, but all models have gaps.
Subscribe for real-time topic updates and unlimited access to our intelligence platform.