DeepSeek is racing to develop its own AI chips after spending a year recruiting engineers and meeting with hardware partners, aiming to break free from dependence on both Nvidia and Huawei as Chinese and American AI giants alike pursue custom silicon for greater control over their technology.
A new study reveals that using JSON mode for AI agents causes smaller models to fail up to 60% of the time and costs up to 11x more than traditional CLI arguments due to shell escaping failures, with developers urged to avoid rewriting CLIs for agents until running measurable experiments.
Shepherd launches an alpha runtime that makes AI agent executions reversible using Git-like forking, delivering 5x faster copy-on-write performance than Docker commit, 95% KV-cache reuse on replay, and OS-level sandboxed permissions that block unauthorized writes before they happen.
Guidance, an open-source Python library with over 21,600 GitHub stars, gives developers precise control over AI model outputs using regex, context-free grammars, and JSON schema validation, while boosting efficiency by reducing unnecessary model processing across backends like OpenAI and llama.cpp.
European AI startup UMA unveils a humanoid robot powered by Real-Time Learning architecture that watches and learns from human demonstrations, positioning Europe as a serious contender in the global robotics race against the US and China.
China is weighing a major policy reversal by restricting foreign access to its most advanced AI models, holding talks with Alibaba, ByteDance, and Z.ai, as U.S.-China AI tensions escalate with accusations of illegal model distillation and retaliatory bans on American AI tools.
Samsung Electronics posts a jaw-dropping 19-fold surge in quarterly operating profit to $58 billion, yet shares plunge over 10% and trigger a market circuit-breaker in Seoul as sky-high investor expectations, fueled by a 150% stock rally tied to AI chip demand, go unmet.
DoorDash launches DashBench, a rigorous AI code review benchmark that replays historical pull requests to surface real developer findings, revealing that multi-stage architectures dramatically outperform single-pass models and that trustworthy AI evaluation demands triangulating multiple imperfect signals rather than relying on any one metric.
OpenAI's GPT-Image-2 dominates the Text-to-Image Arena leaderboard with a record Elo score of 1385, defeating 71 rival models across 5.6 million community votes, with Meta and Reve claiming second and third place respectively.
Broadcom shares surge over 6% after Apple extends their chip partnership through 2031, with Broadcom set to supply custom AI-focused ASIC chips across multiple Apple product generations, including advanced AI servers planned for 2027.