与「C_C」相关的搜索结果

I have been following this guy's content for 4 years. He's not an EV guy at all. Just listen to the great things he has to say about the @cybertruck and Self-Driving: @ Brandon_c_clark on Instagram

It’s #SunDay#! Here’s your space weather report for the week of July 17 - July 23: • 5 M-class flares • 2 C-class flares • 31 coronal mass ejections • 0 geomagnetic storms This video from NASA’s Solar Dynamics Observatory (SDO) shows the week’s activity. It was a fairly calm week on the Sun as SDO’s daily Earth eclipses punctuate our view of the Sun. The spacecraft’s Earth eclipse season, which began on July 5, continues through August 2. Eclipse seasons are three-week periods twice a year when the spacecraft’s orbit passes “behind” Earth, briefly blocking out the Sun each day. Learn more about space weather:

It’s #SunDay#! Here’s your space weather report for the week of July 10 - July 16: • 1 M-class flare • 3 C-class flares • 29 coronal mass ejections • 0 geomagnetic storms This video from NASA’s Solar Dynamics Observatory (SDO) shows the week’s activity. Those “blackouts” in the video below aren’t missing data — those are Earth eclipses! SDO is currently in its 33rd Earth eclipse season, which began on July 5 and will continue through August 2. Eclipse seasons are three-week periods twice a year when the spacecraft’s orbit passes “behind” Earth, briefly blocking out SDO's view of the Sun each day. Learn more about SDO and visualize its orbit:

It’s #SunDay#! Here’s your space weather report for the week of July 3 - July 9: • 1 X-class flare • 36 M-class flares • 3 C-class flares • 42 coronal mass ejections • 1 geomagnetic storm This video from NASA’s Solar Dynamics Observatory (SDO) shows the week’s activity. The Sun celebrated the fourth of July with a big bang, emitting an intense X-class solar flare ( and a burst of M-class flares. You can see the show starting at about 0:49 in the video. Learn more about space weather:

The upcoming wave of SpaceXAI Grok updates is insane Grok 4.5: The 1.5T foundation model is being refined almost daily, and its context window is expected to jump to 1M tokens, possibly as soon as next week Next model: The 2T foundation model is expected to finish training this month and become available to customers next month Grok Build: The harness is getting refinements almost every day based on user feedback. New features, fixes, and workflow improvements are landing at ridiculous speed Inference: SpaceXAI is working on internally developed C/C++ inference mapped directly to GB300 hardware, which could double speed or more Agentic Imagine: Grok will be able to call Imagine as a tool in agentic mode for image and video generation That is going to be huge for game developers, designers, builders, and anyone creating interactive worlds But the biggest thing is this: Grok is being built for real-world engineering Grok 4.5 early beta access was given to Tesla and SpaceX engineers, with feedback coming from real engineering teams solving real-world problems Next month’s release is expected to be another massive step-change improvement as SpaceXAI closes the loop on real-world engineering problems at Tesla, SpaceX, Neuralink, and Boring Company And from now through the end of 2026, Elon said a new from-scratch foundation model is expected to ship every month

AI 视频剪辑 Skill 分享「video-use」 @browser_use 团队推出的开源 Skill，定位为面向 AI Coding Agents（Codex、Claude Code、Cursor、Hermes Agent 等）的视频剪辑 Skill。它不做传统意义上的 Premiere / CapCut 替代品，它是一套让 LLM 通过 “阅读转写文本 + 按需可视化” 来理解视频、并调用 ffmpeg 等工具完成剪辑的 prompt-engineering + 工具脚本集合。 # 核心思想：LLM 不“看”视频，它“读”视频 第一层：音频转写文本（always loaded） 通过 ElevenLabs Scribe 获得逐词时间戳、说话人分离、音频事件标记（如笑声、叹息、掌声），打包成约 12KB 的 takes_packed.md。这是 LLM 的主要“阅读材料”。 第二层：视觉时间线视图（on demand） 仅在决策点（歧义停顿、重拍对比、切点校验）调用 timeline_view.py 生成胶片帧 + 波形 + 字幕的 PNG 复合图。 对比朴素方案“30000 帧 × 1500 tokens = 4500 万 tokens 噪声”，项目走的是 “12KB 文本 + 少量 PNG” 的轻量化路径。这与 Browser Use 让 LLM 读结构化 DOM 而非直接看截图的思路一致。 # 技术流水线：Transcribe → Pack → Reason → EDL → Render → Self-Eval 1. 转写 - transcribe. py / transcribe_batch.py 提取 16kHz 单声道音频，调用 ElevenLabs Scribe，缓存为 transcripts/.json 2. 打包 - pack_transcripts.py 将逐词 JSON 合并为按 0.5s 静音或说话人切换断句的 takes_packed.md 3. 决策 - LLM 自身 阅读 packed transcript，必要时用 timeline_view.py 可视化 4. 生成 EDL - subagents 输出 JSON 格式 edl.json，包含源文件、切点、节奏标签、引用、原因 5. 渲染 - render. py 分段提取 → 无损 concat → 叠动画 → 压字幕 → 响度标准化 6. 自评估 - timeline_view.py + LLM 在输出文件的每个切点 ±1.5s 检查跳帧、爆音、字幕遮挡，最多 3 轮 # 关键工程细节： ffmpeg 为主的剪辑实现 1. 分段提取 + -c copy 拼接（避免叠 overlay 时二次编码） 2. 每段边界 30ms 音频淡入淡出（消除切点爆音） 3. overlay 使用 setpts=PTS-STARTPTS+T/TB 进行时移，确保动画第 0 帧对齐输出时间线 4. 字幕始终最后叠加（防止被动画遮挡） 5. Master SRT 使用输出时间轴偏移：output_time = word.start - segment_start + segment_offset 6. 切点必须落在词边界，并加 30–200ms 填充以吸收 Scribe 50–100ms 的时间戳漂移 7. HDR 源自动 tone-map（HLG/PQ → Rec.709 SDR） 8. 竖屏源自动按高度缩放 9. 两-pass loudnorm：-14 LUFS / -1 dBTP / LRA 11，符合主流社交平台标准 # 动画与包装：多引擎并行 1. HyperFrames：HTML/CSS/GSAP compositions，适合产品 UI、网页转视频、动态排版 2. Remotion：React 组件化 compositions 3. Manim：数学/技术/3Blue1Brown 风格解释动画 4. PIL + PNG sequence + ffmpeg：简单卡片、计数器、打字效果 # SKILL.md 的 12 条“铁律”：生产正确性优先 1. 必须遵守的 12 条硬规则：字幕最后、分段提取再拼接、30ms 淡入淡出、PTS 时移、SRT 输出时间偏移、不切在词中、切点填充、逐词 ASR、缓存转写、并行动画、先确认策略再执行、输出在 /edit/ 2. 其余全部是可调整的“worked example”：调色风格、字幕分块、动画时长、节奏等都可按材料和用户品牌定制

@aaronburnett Truly massive gains will come in ~3 months when the entire training and inference stack is written in C/C++ and massively simplified (most software layers will be deleted completely) and we exact-map Grok to work incredibly well on a GB300