什么是 xAI？

A former Nvidia researcher explains how xAI built Grok Imagine in three months, revealing the training pipeline of video generation models, the definition of world models, and the future trends of Video Agents.

Grok 5 remains under training as of May 2026, with Musk stating it may be indistinguishable from AGI and a ~10% chance of human-level intelligence; XAI built Colossus 2—a 1-GW cluster with ~500k GPUs in 122 days—while facing environmental lawsuits and losing 10 of 11 co-founders.

Cursor订阅用户可保留原计划或切换至新X订阅，新计划不含第三方模型额度，使用Claude/GPT可能更贵。

Grok Imagine Video 1.5 Lite模型现已在AI Gateway上线，支持文本到视频生成，分辨率最高达1080p，时长1-15秒。

Grok 4.6在生物安全监控和对抗性生物任务中表现出色，能有效检测危险查询并允许有益科学查询。

Grok Bot通过五支团队实践构建出Agent操作系统，每个Bot具备长期记忆和工具连接能力，重点解决多任务协作与权限控制问题。

xAI发布的Grok 4.6模型在多个基准测试中表现优异，尤其在知识工作和编码任务上超越了OpenAI和Anthropic的模型。

SpaceX计划将AI业务收入在下月超过其他所有业务，转型为AI公司，星舰将作为AI基础设施的一部分。

xAI的Grok Build CLI工具被曝存在严重安全漏洞，将用户Git仓库和敏感信息上传至Google Cloud存储桶，且未公开数据范围及处理方案。

AWS正式上线xAI的Grok 4.3模型，支持100万token上下文和可配置推理，性能超越多个行业基准。

Grok Build自发布以来全面实施零数据保留（ZDR），默认关闭数据上传并删除历史数据，提供开源工具增强用户隐私保护。

xAI开源Grok Build代码库，回应数据隐私争议，强调用户数据删除与隐私保护。

xAI 官方 Grok CLI 被曝静默上传用户代码库和密钥，引发开发者安全担忧。

AI编码助手迁移旧代码、Grok CLI上传敏感数据、Mesh LLM实现去中心化推理成技术焦点，揭示AI工具链安全与效率新挑战。

xAI团队通过拆分Prefill和Decode、MoE专家分片及通信计算重叠，实现Grok在10万张GPU上的高效部署。

Cursor被SpaceX以600亿美元收购，代码质量将成为软件工程的新瓶颈。

xAI 的 Grok TTS 模型在人类语音相似性测试中得分高达 96，接近真实人类语音。

AI模型增大未必提升性能，新研究发现模型规模与推理能力之间存在非线性关系。

xAI and Cloudflare partner to bring Grok models to Cloudflare AI Gateway, allowing users to pay directly through Cloudflare without additional authentication, environment, or API keys.

Grok Build provides fully free unlimited APIs, making it the easiest way to use Grok Build CLI for 100% free.

The trial has reached a critical stage, with the jury set to decide who is more credible between Musk and Altman.

AI inference will be divided into two types: human-in-the-loop 'answer inference' and agent-based 'agentic inference', with the latter dominating future markets.

Hermes supports configuring multiple domestic and international AI models, including GPT-5.5, Grok-4.3, Gemini, etc.

AI chip demand surges, Cerebras raises IPO price range to $150-$160, GPU dominance in AI computing shifts towards heterogeneity.

xAI and Anthropic have reached a data center deal, sparking discussions about environmental impact and AI ethics.

AI video agents will follow the same trajectory as coding agents, with Grok Imagine achieving a zero-to-one breakthrough through real-time interactive world models and generative UI, leading to future video generation driven by intelligent agents with cameras, editors, and tool belts rather than text prompts.

AI video generation is following a similar evolution path as coding agents, with Grok Imagine demonstrating a leap from zero to one; future systems will evolve into interactive agents with cameras, editors, and tool belts, rather than simple prompt boxes.

Third Point founder Dan Loeb argues AI is reshaping investment logic: investors must grasp the AI Stack (power → chips → models → applications); credit expertise is critical in volatile markets; his $25B multi-asset platform coordinates equities, credit, insurance, and private markets via strategies like Fulcrum Security; activist investing succeeds through governance reform, not confrontation.

AI has not triggered the predicted 'job apocalypse'; OpenAI’s Sam Altman and Anthropic’s Dario Amodei recently revised their earlier dire forecasts, admitting their job-impact predictions were overly pessimistic; in practice, AI boosts individual productivity but does not reduce total workload—instead enabling more high-leverage tasks and new roles.

The article analyzes the top five labs in Text Arena rankings and their models, showcasing the distinct strengths and tradeoffs of frontier models in different fields. AnthropicAI's Claude Opus 4.7 is the most comprehensive, while Google DeepMind's Gemini 3.1 Pro excels in creative writing.