Gemini

JetBrains applied Google DeepMind’s AlphaEvolve to optimize the deeply tuned B-tree indexing core in IntelliJ IDEs, achieving 15–20% synthetic performance gains and validating two candidates that reduced end-to-end indexing time from 17.4s to under 16.8s in real IDE integration tests.

SaaS-Bench evaluation shows mainstream large models have less than 4% complete pass rate on real office tasks, revealing huge challenges for AI fully automated office work.

AI automation replacing human workers destroys its own customer base; tech elites promote UBI not out of charity but as a modern form of forced consumption to sustain subscription economies, mirroring 19th-century indentured labor systems after slavery abolition.

HTML-in-Canvas 是一项突破性 Web API，让 HTML 元素原生嵌入 Canvas 并参与布局与交互，同时支持 GPU 着色器渲染、3D UI 贴图和无障碍访问，彻底解决传统 Canvas 的语义缺失痛点。

Google introduces its AI Threat Defense system, which uses automation and prioritized threat handling to help organizations combat advanced AI-driven attacks.

Global LLM rollouts eliminate control groups, invalidating naive before/after comparisons; this tutorial introduces synthetic control in Python to build counterfactuals using weighted combinations of untreated units.

DeepMind创始人Demis Hassabis在播客中分享了从国际象棋神童到诺奖得主的历程，探讨了通用人工智能（AGI）的关键缺失组件，如持续学习与长期推理，并揭示了AlphaFold、Gemini模型对科学发现的影响及未来AI在材料科学、药物发现等领域的变革潜力。

Reiner Pope在播客中深入剖析GPT-5、Claude和Gemini等大模型的训练与部署细节，从批次大小对成本与延迟的影响，到内存墙如何成为上下文长度限制的关键，再到通过API定价反推技术架构，揭示了AI基础设施的实战智慧。

Assembly AI shipped major advancements in May: LLM Gateway now supports reasoning via a single parameter (low/medium/high) and abstracts provider differences; streaming speaker diarization accuracy jumped with 66% fewer false alarms and 60% fewer phantom turns, per-word speaker labels, and continuous partials every ~3 seconds; Playground now previews 34 voices and allows public sharing; PII redaction is live for streaming, automatically removing names, phone numbers, credit card numbers in real

NVIDIA introduces the 550B-parameter Neotron 3 Ultra, a mixture-of-experts agent model trained for task orchestration, outperforming many trillion-parameter open agents on benchmarks, with full data and recipe transparency to enable enterprise on-prem deployment and fine-tuning.

Li Fei-Fei's team releases ESI-Bench evaluation benchmark, transforming observers into actors for the first time, closing the perception-action loop, and discovering three major issues in current AI spatial intelligence: action blindness, 3D reconstruction traps, and metacognitive deficits.

Google Cloud API Gateway新增流式传输支持，允许实时和AI应用在不改变现有规范的情况下使用，提升API管理的灵活性和AI集成能力。

Google 推出 Agent Compute Engine（ACE）和扩展治理功能，通过微虚拟机、内存共享等技术实现 Gemini 代理的高效安全运行。

Google推出Agent Compute Engine，为每个AI代理提供硬件隔离的微虚拟机，解决agentic AI在大规模部署中的性能与安全挑战。

Google Cloud 现在提供 OpenAI 模型，增强企业 AI 平台的灵活性和选择性。

Google DeepMind利用Gemini代理在数学发现领域取得突破，自主发现并验证了Busy Beaver问题中Turing机器的加速规则。

Google开源Harness²框架，通过双层递归机制解决AI代理在无验证器环境下的自我优化难题，实现跨模型迁移性能提升。

Google Cloud 提出 SAVE 方法，通过分离角色、验证事实和提前审核，提升生成媒体的可信度。

构建代理插件的四个评估工程教训：评分执行步骤、设计异步评估、使用工具调用轨迹、测试边缘情况。

Google Research推出Geospatial Reasoning智能体，通过整合地球AI基础模型和空间工具，实现跨领域地理分析与行星级预测，已与WHO等机构合作验证。

Google Gemini为中小型企业提供了一体化的AI代理工具，提升效率并控制成本。

Docker Agent通过声明式YAML配置实现AI代理容器化，支持多模型协作与OCI仓库分发，已获3300星标。

Google推出Gemini代理，作为集成知识工作、代码生成、内容创作的通用工作代理，支持在Gmail/Drive等工具内直接操作，并提供行业专用工具和安全治理方案。

A16Z报告指出AI使用广泛但深度不足，头部1%用户月均花费903美元。

谷歌提出AI视频共导演框架，通过多智能体协作解决长视频生成中的语义漂移和级联故障问题，实现分钟级连贯视频生成。

Google通过Gemini AI和差分模糊测试将关键C依赖库迁移至Rust，消除内存漏洞并实现ABI兼容。

Google AI订阅者可使用Gemini语音功能搜索Gmail、管理Keep任务并创建Docs，Docs Live功能尤具价值。

Stripe通过构建内部AI代理Kai，展示了如何在企业中实现安全、可扩展的AI应用，涵盖治理、数据层和技能平台。

Google在Gemini中推出动态视频理解功能，减少88% token和66%成本。

Google DeepMind推出Gemini模型的agentic视频理解功能，可减少88%的token消耗，降低成本66%，并提升7%准确性。