与「idle 」相关的搜索结果

Grok Build just got another meaningful upgrade, bringing permanent session deletion, smarter diagnostics, better keyboard guidance, and stronger reliability across session management, background tasks, and compaction Release Notes: v0.2.118 Features: • Sessions can now be permanently deleted from the dashboard by pressing Ctrl+X twice on an idle row, or from the welcome list with d then y. • Keyboard shortcuts help (Ctrl+.) now shows how to browse prompt history and search the conversation. • grok doctor now warns when tmux is reducing colors and can fix the config. Bug Fixes: • /btw now retries on temporary model overload instead of failing immediately. • Session sharing is temporarily disabled. • [stop] / Ctrl+C during /compact now cancels instead of no-opping. • Automatic recaps no longer appear twice after the same turn. • Background task wait timeout descriptions and limits now match the client's actual configured ceiling. • Background tasks no longer stay stuck as 'Running' in the tasks pane when they finish quickly. • Plan mode indicator now disappears right after approving a plan instead of lingering. • Dragging the scrollbar in the plan preview now works as expected. • Compaction now correctly handles certain context-length errors from the inference API.

Your Money Made Smarter. Your stablecoins can do more than sit idle. With RWUSD, earn 3.36% APR, paid daily, with rewards benchmarked against tokenized U.S. Treasury Bills. Binance it 👉

Former $CRWV employee on why neoclouds are far more exposed to GPU generation cycles than hyperscalers ( $MSFT, $AMZN, $GOOGL ): - The expert describes GPU utilization tracking at hyperscale as a continuous and disciplined process built around two lenses. The first is infrastructure utilization, covering GPU occupancy, idle time, and memory utilization, noting that 95% booked usage can still mask inefficiency if jobs stall or batches have idle gaps. The second is outcome utilization, asking whether the compute is actually generating business value, measured by metrics such as tokens trained per dollar, time to reach target accuracy, and tokens per second per GPU. - The expert sees a meaningful difference between hyperscalers and neoclouds on GPU investment economics. Hyperscalers like $MSFT, $GOOGL, and $AMZN can tolerate a 3-5 year payback period given their ability to monetize the same infrastructure across multiple revenue streams. Neoclouds like $CRWV operate on a tighter 2-3 year window. - With higher financing costs and direct dependence on infrastructure cash yield, neoclouds are far more sensitive to utilization and GPU residual risk. The biggest risk is GPU generation cycles, where a slow payback means newer chips could erode pricing power before the asset has paid itself off. - The expert explains that for hyperscalers, roughly 60% of GPU capacity is allocated to external monetization, including GPU rentals, managed AI services, enterprise inference workloads, and startup model training. The remaining 40% is used internally, and of that internal portion, the majority is still indirectly monetized through products like M365 Copilot or GitHub. Around 40% is dedicated to pure R&D. - The GPU pricing mix has shifted meaningfully over the past few years. In 2023, around 70-80% of revenue was hourly as customers paid a premium just to get access to scarce GPUs. By 2025 that had moved to roughly 50% hourly and 30-50% committed, and the expert expects 2026 to tip further toward committed at around 65% for hyperscalers as AI matures and inference becomes more predictable. Neoclouds are moving in the same direction but more slowly. - By 2027-2030, the expert sees committed contracts settling at 55-65% as the norm, with hourly pricing remaining but losing its scarcity premium as more supply comes online.

Grok Build just got another major update, giving developers deeper visibility into token usage and costs, more flexible model configuration, smarter prompt batching, and enhanced diagnostics for troubleshooting Release Notes: v0.2.109 Features: • /usage now shows token counts and cost for the current session. • grok doctor fix terminal.ssh-wrap can install the recommended SSH wrapper alias. • [model_providers.] lets operators share gateway settings across custom models. • Reasoning effort now accepts max as its own tier (above xhigh) when the model advertises it. • Queued follow-ups can now be batched into a single model turn with the new combine_queued_prompts setting. • /doctor is now the main slash command for terminal, tmux, clipboard and keyboard diagnostics. • read_file now returns full Markdown files inside skills/directories without truncation. Bug Fixes: • Voice dictation now explains when the microphone delivered only silence (macOS permission) versus no speech detected. • Duplicate 'Worked for' markers no longer stack in the transcript when background tasks defer during a parked turn. • The idle status row now clearly says '1 subagent still running' instead of 'watching · 1 subagent' when background work remains. • Background /loop iterations no longer overlap when descendant subagents are still running.

As we celebrate 9 years of Binance, we're continuing to build new ways to put your assets to work. VIP Exclusive, now live. Unlock up to 4.12% APR on your idle stablecoins with RWUSD. Your Money Made Smarter. 👉

VIP Exclusive. Now live. Unlock up to 4.12% APR on your idle stablecoins, benchmarked against tokenised US T-bills. VIP status isn’t just about fee rebates, it’s about your capital working harder for you. Learn more / T&Cs 👉

他也补充列出了一些此前遗漏的公司： 数据市场： Kled、Luel 电脑 / 浏览器使用： Chakra Labs、Habitat、Refresh、Cua、Originator、Dojo 编程 / 软件工程： Proximal、Idler、Calaveras、BenchFlow、Vmax 长程任务 / 推理： Andromede、Aviro、AIChamp、General Reasoning、Champ、Haladir、Hillclimb 模型行为 / 评测： Gray Swan、Theta、Preference Model、Vals AI、Andon Labs、Verita、Good Start Labs 垂直领域： Halluminate、Quesma、Phinity、Rise Data Labs、Kairos、TrainLoop、Huzzle Labs

用一张人物图，做成 Codex 专属动画宠物：省 token 版教程 最近我用一张人物形象图，做了一个能放进 Codex 宠物库的自定义宠物。但花费了我五小时80的额度，我就把最省token的教程做出来了 重点是：不要让模型反复出草稿、不要生成 16 张独立方向图，否则时间和 token 都会迅速膨胀。 第一步： 准备一张参考图 建议使用：人物清晰、最好是正面或 3/4 视角、尽量全身、发型和服装明确、背景不要太复杂。 第二步： 1.在codex中找到个人信息 2.设置 3.宠物 4.创建 直接复制这段提示词 使用 hatch-pet 技能，把这张人物图做成 Codex v2 自定义动画宠物。 请使用省 token 工作流： 1. 只生成 1 张主形象作为身份基准； 2. 生成 idle、running-right、waving、jumping、failed、waiting、running、review； 3. running-left 从 running-right 逐帧镜像生成，不要重新出图； 4. 用 1 张上右下左的四方向基准图； 5. 用 2 条八格方向条完成 16 个视线方向，不要生成 16 张独立单图； 6. 不要生成草稿或多个候选版本； 7. 每次生成只返回文件路径和一句质检结果，不要回传图片预览； 8. 最终安装到 ~/.codex/pets/ 第三步、放进 Codex 宠物库 打开 Codex 的宠物选择页，点击右上角刷新按钮，找到宠物后点击“选择”。如果刷新后还没显示，重启一次 Codex 再刷新。 一句话总结 想快、想省 token，就记住：一张主形象 + 八组动作 + 一组四方向基准 + 两条八格方向条；向左移动直接镜像，不要生成草稿和 16 张独立方向图。