Meta 的開源本地智能體模型面臨隱私悖論
meta 發布了 muse glimmer，這是一個 30b 參數的 apache 2.0 開權本地智能體模型，標誌著向端側“個人超級智能”的轉變，並界定了開源與專有 ai 的界限。
meta unveiled muse glimmer, a 30-billion parameter open-weight model available under the apache 2.0 license, specifically engineered to execute complex, multi-step agent workflows like coding and file management locally on macs or pcs. unlike its more powerful closed counterpart, muse spark, glimmer supports text and image processing across 100 languages and functions offline to ensure user data remains on-device. this release delineates meta's hybrid approach: distributing powerful, customizable tools for widespread personal use while retaining control over its most advanced closed-weight capabilities. by enabling 'always-on' local agents, meta challenges the cloud-dependent ai dominance of competitors like openai and google, betting that privacy and ownership will drive the next wave of ai adoption.