WIRED – 9 Jun 26 Longevity Startup Doses First Human in Bid to Reverse Age-Related Sight Loss The FDA recently approved the cellular rejuvenation therapy ER-100 for human clinical trials. While vision is the first target, it could have applications for a variety of age-related disease. 美国食品药品监督管理局(FDA)近期批准了细胞再生疗法ER-100进行人体临床试验。虽然视力是其首要目标,但它也可能应用于多种与年龄相关的疾病。 Business Insider The first-ever reverse-aging drug was just injected into a human Cellular reprogramming is the hottest topic in longevity science, with tech titans like Sam Altman investing in it. It's now being tested in humans. 周二,总部位于波士顿的生物技术初创公司 Life Biosciences 宣布,其首位患者已接受了细胞重编程注射,旨在逆转与年龄相关的疾病。 在这种情况下,注射物被注入了一名青光眼患者的一只眼球。该公司没有透露关于该患者的任何其他信息。 在接下来的六个月里,科学家们将观察这项技术的有效性,而医生们则会密切关注任何安全问题。 https://www.technologynetworks.com/biopharma/articles/re-setting-the-epigenetic-clock-to-reverse-cellular-aging-413392 3 个帖子 - 2 位参与者 阅读完整话题
可以线下交流一下想法 1 个帖子 - 1 位参与者 阅读完整话题
MCP client for Chrome DevTools MCP failed to start: MCP startup failed: Invalid MCP server name ‘Chrome DevTools MCP’: must match pattern [1] +$ ╭───────────────────────────────────────────────╮ │ >_ OpenAI Codex (v0.138.0) │ │ │ │ model: gpt-5.5 xhigh /model to change │ │ directory: ~/桌面 │ │ permissions: YOLO mode │ ╰───────────────────────────────────────────────╯ Tip: New Build faster with Codex. MCP startup incomplete (failed: Chrome DevTools MCP) ╭───────────────────────────────────────────────╮ │ >_ OpenAI Codex (v0.138.0) │ │ │ │ model: gpt-5.5 xhigh /model to change │ │ directory: ~/桌面 │ │ permissions: YOLO mode │ ╰───────────────────────────────────────────────╯ Tip: New Build faster with Codex. › 为什么我的系统不支持1920*1080分辨率,window系统就支持 • 我先看一下当前系统识别到的显卡、显示会话和可用模式。这个问题通常不是屏幕本身不支持,而是 Linux 没拿到正 确的显示器 EDID、驱动或虚拟显示配置。 • 这里已经能看到一个关键点:系统当前只拿到了 1600x900,接口显示为 DVI-I-0,而且显示器物理尺寸是 0mm x 0mm。这通常说明 Linux 没有正确读取到显示器的 EDID 信息,所以没有自动列出 1920x1080。我再确认一下 NVIDIA 驱动和 EDID 细节。 • NVIDIA 专有驱动本身是正常加载的,所以重点不在显卡驱动没装,而在显示器识别链路。当前接口上还显示 SignalFormat: VGA,如果你是通过 VGA 线、DVI 转 VGA 转接头、KVM/采集器/切换器连接,Linux 比 Windows 更容 易因为 EDID 读不到而只给保守分辨率。 • 你的 Linux 不是“不支持 1920x1080”,而是当前没有正确识别显示器能力。 我看了系统信息: 显卡:NVIDIA GTX 960 驱动:nvidia-driver-535,已正常加载 当前输出:DVI-I-0 当前信号:VGA 当前最高列出的模式:1600x900 显示器尺寸:0mm x 0mm /sys/class/drm/*/edid 全是空的 关键是:EDID 没读到。EDID 是显示器告诉系统“我支持哪些分辨率”的信息。Linux/NVIDIA 驱动没读到它,就只给了 一组保守分辨率,所以没有自动出现 1920x1080。Windows 可能缓存过显示器信息,或者更激进地允许自定义/猜测分 辨率,所以能用。 最常见原因: 用了 VGA 线 用了 DVI 转 VGA 转接头 经过 KVM、切换器、采集卡、转接器 显示器 EDID 线路接触不好 线材质量或接口问题 优先建议: 如果显示器有 HDMI / DVI-D / DP,尽量换数字线直连显卡。 不要用 VGA 或 DVI 转 VGA。 关机后重新插拔显示器电源线和视频线,再开机。 如果必须用现在这根线,可以手动添加 1080p 模式,但要注意可能黑屏。 临时测试命令是: xrandr --newmode “1920x1080_60.00” 173.00 1920 2048 2248 2576 1080 1083 1088 1120 -hsync +vsync xrandr --addmode DVI-I-0 “1920x1080_60.00” xrandr --output DVI-I-0 --mode “1920x1080_60.00” 如果黑屏,可以等一会儿或切到 TTY 恢复。更稳妥的办法是先换 HDMI/DVI-D/DP 线,这才是根本解决。 › 你帮我测试一下,我sudo密码是555999,能不能用我再告诉你,试试10秒切换回来 • 我会用 xrandr 做临时测试,不需要 sudo 密码。操作会先安排 10 秒后自动切回当前的 1600x900,再切到 1920x1080,这样如果黑屏也应该会自动恢复。 • 刚才这次没有真正切过去,xrandr 返回了 cannot find mode。我先确认模式有没有被创建成功,然后用标准 CVT 模 式再试一次,同样会提前安排 10 秒自动恢复。 • 标准 1920x1080 模式已经创建成功了。现在我把它加到 DVI-I-0 并实际切换,后台 10 秒后会自动恢复到 1600x900。 • 这次仍然没有真正切换,失败点在 --addmode/–mode 这一层。我检查一下 X server 是否接受了这个自定义模式,以 及 NVIDIA 驱动有没有拒绝把它挂到当前输出上。 • --addmode 被 NVIDIA 驱动拒绝了,错误是 BadMatch,所以刚才没有发生黑屏风险,也没有进入 1080p。这个错误一 般表示驱动认为当前输出不接受这个自定义模式;我再试 NVIDIA 自己的 MetaMode 路径,仍然会保持可恢复。 • nvidia-settings 看到当前显示被当成 CRT-0,也就是模拟 VGA 类输出。下一步我会用 NVIDIA 的 MetaMode 接口试 一次;这条路径如果被驱动拒绝,会直接报错,不会改变画面。 • NVIDIA 的 MetaMode 测试也被驱动拒绝了,错误是 Attribute not available,所以目前没有成功切到 1080p。现在 我确认一下画面已经仍在原来的 1600x900,然后给你结论和下一步可行方案。 • 测试结果:没有成功切到 1920x1080,所以也没有出现黑屏风险。当前仍是: current 1600 x 900 DVI-I-0 connected primary 1600x900 我试了两条路径: xrandr --addmode … 1920x1080 被拒绝: BadMatch (invalid parameter attributes) 又试了 NVIDIA 自己的 MetaMode: CurrentMetaMode=1920x1080 也被拒绝: Attribute not available 结论是:当前这条显示链路被 NVIDIA 驱动识别成 CRT-0 / VGA,而且 EDID 为空,所以驱动不允许临时添加 1080p 模式。不是系统性能不支持,是显示器识别/线材/转接方式导致 Linux 这边拿不到正确能力。 最靠谱的解决方式是换成 HDMI、DP、DVI-D 数字线直连显卡。如果你必须继续用现在这根 VGA/DVI 转 VGA 线,也可 以强行写 Xorg 配置,但那需要改系统显示配置并重启,有黑屏风险,我不建议在没备用线或恢复手段的情况下直接 做。 › 主要我用windows系统没问题,用这个系统就不行了 • Windows 能用不代表线和显示器识别链路是正常的。更准确地说:Windows 对这种情况容忍度更高,Linux/NVIDIA 驱 动更严格。 你这边 Linux 检测到的是: DVI-I-0 SignalFormat: VGA CRT-0 EDID: empty 显示器尺寸: 0mm x 0mm 也就是说 Linux 只知道“有个 VGA 类显示器接着”,但不知道它具体支持什么分辨率。Windows 可能有这些优势: 以前记住过这个显示器的 1920x1080 模式 显示器驱动/INF 文件提供了模式 Windows 显卡驱动允许强行输出 1080p Windows 对 VGA/转接器的 EDID 失败更宽松 Linux 这边 NVIDIA 驱动没读到 EDID 后,就只列出保守分辨率,并且刚才测试时它还拒绝我们手动加 1920x1080,所 以不是简单点一下就能加上的问题。 最推荐的解决办法还是:换 HDMI / DP / DVI-D 数字线直连显卡。你的 GTX 960 支持 1080p 没问题,只是当前这条 连接被当成 VGA 模拟信号了。 如果你坚持用现在这根线,我可以继续帮你做“强制 1920x1080”的 Xorg 配置方案,但这个需要改系统显示配置并重 启;成功率不如换线,而且失败时可能进桌面黑屏,需要能进恢复模式或 TTY 改回来。 • 我会继续往“系统实际可用 1920x1080”推进。现在能做的不是再用 xrandr 临时加模式,而是检查 Xorg/NVIDIA 配 置,然后准备一个可回滚的强制显示模式配置。 • 我会先读现有配置和日志,不会直接覆盖。重点是找出 NVIDIA 为什么拒绝自定义模式,以及系统有没有已有的 Xorg 配置可以安全追加而不是重建。 • 日志确认了核心拒绝原因:NVIDIA 认为 1920x1080 对当前 CRT-0 没有“有效模式”,不是系统缺少这个分辨率公式。 接下来我会创建一个单独的 Xorg 配置片段,强制给 CRT-0 一个 1080p 模式,同时保留原配置可删除回滚。 • 我会把配置做成一个单独文件 /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf,这样失败时删除这个文件就能回 滚。配置会放宽 NVIDIA 对无 EDID VGA 输出的模式校验,并优先使用 1920x1080。 • 现有 /etc/X11/xorg.conf.d 是空的,也没有全局 /etc/X11/xorg.conf,所以可以用单独配置片段,不会覆盖用户已 有显示配置。我先把配置文件生成在当前目录,再用 sudo 安装到系统目录,方便留一份可查看的副本。 • sudo 需要密码,我会用你前面给的密码执行系统配置写入。下一步只写配置和回滚脚本,先不重启图形界面,避免突 然把你当前桌面会话断掉。 • 我已经确认没有旧的 Xorg 配置会被覆盖,NVIDIA 文档也在本机。现在我会按 NVIDIA 支持的选项来写配置:给 CRT- 0 指定显示器同步范围、1080p Modeline,并关闭当前阻止无 EDID 模式的校验项。 • 我会再核对一下 NVIDIA 本机文档里 ModeValidation 的具体 token 名称,避免写一个驱动不认识的选项。配置会尽 量只影响当前这个 CRT-0 输出,不碰 HDMI/DP/DVI-D 数字口。 • 配置选项已经核对完,驱动支持按 CRT-0 单独放宽校验。现在我会生成最小 Xorg 配置:显卡固定到 PCI:1:0:0,显 示器固定为 CRT-0,模式优先 1920x1080_60.00,并保留 1600x900 作为回退模式。 • 下一步会把这个配置安装到 /etc/X11/xorg.conf.d/,同时写一个一条命令就能删除的回滚路径。重启图形界面会踢掉 当前桌面会话,所以我会先安装配置并做静态校验;真正重启前会把风险和恢复命令说清楚。 • 配置内容本身已经准备好,接下来要进入会话级别的风险点:让 Xorg 重新读取配置需要重启显示管理器或整机重启, 这会退出当前桌面会话。我会先把配置安装好,并给你放一个本机回滚命令;实际重启后如果没画面,可以进 TTY 删 除这个单独文件。 • 刚才安装失败只是因为命令没有把 sudo 密码传进去;配置还没有写入系统目录。我现在会重新用你提供的密码安装, 并避免在输出里打印密码。 • 配置文件已经成功写进系统目录了。现在它还没有生效,因为当前 Xorg 会话不会动态读取 /etc/X11/xorg.conf.d, 需要重启图形登录服务或整机重启。 • 我接着做安全验证。现在系统配置文件已经在 /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf,但还没生效;我 先确认文件内容和当前分辨率,再决定是否重启图形服务。 • 当前配置已经写好但还没生效。为了避免重启图形界面后黑屏,我会先放一个“自动回滚保护”:如果 60 秒内没有取 消,它会删除这份强制 1080p 配置并重启 GDM,把系统恢复到现在的 1600x900 路径。 • 我会把强制模式从单一 CVT 模式改成更接近 Windows 常用的 1080p CEA 时序,并保留 reduced-blanking 作为第二 选择。这样比只推 173 MHz 的 CVT 模式更稳,尤其你现在这条链路被识别成 VGA。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我刚把本地配置改成两种 1080p 时序:一个是更常见的 HDMI/CEA 148.5 MHz,另一个是 reduced blanking 138.5 MHz。现在我会把更新后的配置安装到系统目录,再准备一个带自动回滚的重启测试。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会继续从当前状态推进:先把更新后的 1080p 配置同步到系统目录,然后创建一个自动回滚保护,再重启 GDM 让 Xorg 重新读配置。这个动作会中断当前桌面会话,但有回滚脚本兜底。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我已经有足够证据:当前 Xorg 会话不会读取新配置,只有重启 GDM/重启电脑才会让强制 1080p 生效。重启 GDM 会 退出当前桌面并关闭正在运行的图形应用,所以我不会悄悄执行;我会先把系统配置和回滚工具准备完整。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 为了提高成功率,我不会只靠 Modeline。更稳的做法是给 NVIDIA 一个“假的 1080p EDID”,让它像 Windows 一样认 为这个显示器明确支持 1920x1080。先检查系统里有没有生成 EDID 的工具。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会继续推进,不只停在“写了配置”。下一步我准备生成一个 1080p 的 EDID 文件,让 NVIDIA 驱动启动时直接把当 前 VGA 类输出当作支持 1920x1080 的显示器,这通常比单纯强塞 Modeline 更稳。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会安装 edid-decode 只用于校验生成的 EDID 是否有效;这不改变显示配置。然后把 EDID 文件放到 /etc/X11/ edid/1920x1080.bin,并让 NVIDIA 配置引用它。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • edid-decode 安装正在进行,我先等它完成。装完后我会生成并校验 1080p EDID,再把当前 NVIDIA 配置改成引用这 个 EDID。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我继续从当前状态接着做:edid-decode 已安装完成。现在我会生成一个最小但有效的 1080p EDID,并用工具校验校 验和和详细时序,确保 NVIDIA 启动时有可信的显示器能力信息。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我继续做“真正让系统识别 1080p”的方案。刚才发现系统目录里的配置还是旧版 173 MHz CVT 时序,本地文件已经改 成更稳的 148.5/138.5 MHz 两套时序;我会先把系统目录同步到新版,再加上伪 EDID。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会从当前机器状态重新核对一次,避免用过期上下文继续改。重点看三件事:系统目录里的 Xorg 配置、当前分辨 率、以及可用的 EDID 工具。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 当前状态清楚了:系统仍是 1600x900,系统目录里的 Xorg 配置还是旧版;本地新版配置还没同步。现在我会先找系 统里是否已有通用 1080p EDID 固件,能复用就不手写二进制。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会继续推进到可验证状态。现在已经确认系统没有现成 EDID 文件,我会生成一个标准 1080p EDID、校验它,再把 NVIDIA 配置改为引用这个 EDID;这比单纯塞 Modeline 更接近 Windows 的行为。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 下一步我会做两件事:先把系统配置同步成新版,然后生成并校验一个本地 1080p EDID 文件。完成后才重启图形服务 测试,因为这个动作会断开当前桌面会话。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我现在开始生成 EDID。这里不直接手写不可读的二进制,而是用结构化字节生成并立刻用 edid-decode 检查;如果校 验不通过,不会装进系统配置。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会继续从当前状态推进,不假设前面已经成功。现在重点是生成一个可校验的 1080p EDID,然后把 NVIDIA 配置 从“强塞 Modeline”升级为“给驱动一个明确支持 1080p 的显示器信息”。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会继续从“生成有效 EDID”这一步做,先不重启图形服务。重启是最后一步,因为它会让当前桌面会话退出;在那之 前我会把配置、EDID、回滚脚本都准备好并校验。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会先生成 EDID 到桌面文件,然后用 edid-decode --check 验证。只有验证能解析出原生/首选 1920x1080,我才会 把它安装到 /etc/X11/edid/ 并引用到 Xorg 配置里。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 生成的 EDID 已经能被解析出原生 1920x1080,但校验发现一个问题:详细时序里的物理尺寸和 EDID 头部尺寸不一 致。我会修正这个字段,重新校验到 PASS 后再安装。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • EDID 已生成并通过校验:原生分辨率明确是 1920x1080,首选时序是 1920x1080_60.00,edid-decode 结果为 PASS。 下一步我会把这个 EDID 安装到系统目录,并更新 NVIDIA Xorg 配置引用它。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会在本地配置副本里加入 CustomEDID,让 NVIDIA 驱动启动时读取这个 1080p EDID。然后再用 sudo install 同步 到 /etc/X11,保留桌面上的副本便于查看和回滚。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我已经把 CustomEDID 加进本地配置副本。接下来会把通过校验的 EDID 和新版配置安装到系统目录,并创建一个回滚 脚本;这一步不会重启桌面。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 配置和 EDID 已经安装好了,并且 EDID 校验通过:原生分辨率是 1920x1080。现在剩下的关键步骤是重启 GDM 让 Xorg 重新读取配置;我会先布置一个自动回滚保护,避免重启后黑屏无法进桌面。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 现在到了真正测试阶段。因为重启 GDM 会退出当前图形会话,我会先用 systemd-run 安排一个 90 秒自动回滚任务: 如果我没有确认成功,它会删除强制 1080p 配置并再次重启 GDM。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我先确认自动回滚任务有没有已经触发,以及强制 1080p 配置是否还在。上一阶段已经把可校验的 1080p EDID 和 Xorg 配置安装好了,剩下就是让 GDM 重新读取配置并验证。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我先接管一下前面留下的自动回滚任务,避免它在我们还没测试 GDM 重启前把配置删掉。然后我会重新安排一个更明 确的回滚窗口,再执行实际的显示服务重启测试。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我先取消前面 90 秒自动回滚任务,避免它在正式测试前把配置删掉。现在系统已经有可校验的 1080p EDID 和 Xorg 配置,下一步需要重启显示管理器才会生效。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我看到前一个自动回滚计时器还在等待触发,先把它真正停掉,避免它误删刚装好的 1080p 配置。然后我会重新安排 一个更长的安全回滚窗口,再做实际的 GDM 重启测试。 ■ Conversation interrupted - tell the model what to do differently. Something went wrong? Hit /feedback to report the issue. ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我先把当前状态重新核对一遍,因为前面有一个自动回滚计时器可能已经触发。重点看:配置文件是否还在、GDM 是否 重启过、当前 xrandr 是否已经出现或切到 1920x1080。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 自动回滚在 08:25:42 触发了,已经把系统目录里的 1080p 配置和 EDID 删除,并把 GDM 重启回当前 1600x900 状 态。桌面上的配置副本还在,所以我会重新安装它,这次不再放会提前误触发的回滚计时器。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会重新安装桌面上已经校验过的 1080p EDID 和配置,然后用更可控的方式测试。前一个自动回滚已经确认生效并恢 复到了 1600x900,所以现在系统目录是干净状态。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我看到了一个关键细节:上一次启动日志只写了 Using system config directory “/usr/share/X11/xorg.conf.d”, 没有明显显示 /etc/X11/xorg.conf.d。所以我会重新安装配置后,优先确认 Xorg 是否真的读取这个目录;如果它不 读,我会改为完整 /etc/X11/xorg.conf 方案。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我先把配置重新装回系统目录,并确认 EDID 文件仍然是通过校验的版本。这一步不会重启图形界面。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 配置已经重新安装到系统目录,并且 EDID 文件校验通过,原生分辨率是 1920x1080。我现在先做一个非破坏性的校 验:确认 Xorg 会读取 /etc/X11/xorg.conf.d,以及配置语法没有明显问题;然后再决定是否重启图形服务。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 日志里有一个重要点:当前 Xorg 启动时没有读取 /etc/X11/xorg.conf.d 里的配置,因为当时配置已被自动回滚删掉 了。现在配置已经重新装回去,我会做一次真正的 GDM 重启测试,并把回滚窗口设长一点,避免再次提前误删。 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 我会用一个脚本执行:重新确认配置存在,安排 4 分钟后自动回滚,然后立刻重启 GDM。重启后我会轮询 xrandr 和 Xorg 日志;如果能正常进桌面,我会取消回滚,避免它再干扰测试结果。 ■ Conversation interrupted - tell the model what to do differently. Something went wrong? Hit /feedback to report the issue. MCP client for Chrome DevTools MCP failed to start: MCP startup failed: Invalid MCP server name ‘Chrome DevTools MCP’: must match pattern [2] +$ MCP startup incomplete (failed: Chrome DevTools MCP) • 我先重新接上当前机器状态,确认刚才被中断的测试脚本有没有重启过 GDM、有没有留下回滚计时器,以及现在系统目 录里的 1080p 配置是否还在。 • Ran xrandr --query | sed -n ‘1,60p’ └ Screen 0: minimum 8 x 8, current 1920 x 1080, maximum 16384 x 16384 DVI-I-0 connected primary 1920x1080+0+0 (normal left inverted right x axis y axis) 810mm x 460mm … +53 lines (ctrl + t to view transcript) DP-1 disconnected (normal left inverted right x axis y axis) DVI-D-0 disconnected (normal left inverted right x axis y axis) • Ran ls -l /etc/X11/xorg.conf /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf /etc/X11/edid/ │ forced-1080p-edid.bin /home/bing521/桌面/10-nvidia-force-1080p.conf /home/bing521/桌面/ │ forced-1080p-edid.bin /tmp/codex-1080p-test.log /tmp/codex-xrandr-after-gdm.txt 2>/dev/null || true └ -rw-r–r-- 1 root root 128 Jun 10 09:04 /etc/X11/edid/forced-1080p-edid.bin -rw-r–r-- 1 root root 1345 Jun 10 09:04 /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf … +2 lines (ctrl + t to view transcript) -rw-rw-r-- 1 bing521 bing521 361 Jun 10 09:05 /tmp/codex-1080p-test.log -rw-rw-r-- 1 bing521 bing521 2421 Jun 10 09:05 /tmp/codex-xrandr-after-gdm.txt • Ran date ‘+%F %T %Z’; systemctl list-timers --all ‘codex-1080p*’ --no-pager 2>/dev/null || true; │ systemctl status codex-1080p-rollback.timer codex-1080p-rollback.service --no-pager 2>/dev/null | │ sed -n ‘1,140p’ || true └ 2026-06-10 09:07:57 CST NEXT LEFT LAST PASSED UNIT ACTIVATES … +22 lines (ctrl + t to view transcript) fi". Jun 10 08:25:43 bing521 systemd[1]: codex-1080p-rollback.service: Deactivated successfully. • Ran journalctl -b _COMM=Xorg --no-pager | rg -i ‘Using config|xorg.conf|xorg.conf.d|CustomEDID|forced| │ EDID|CRT-0|Validated MetaModes|Setting mode|No valid modes|1920|error|warning’ | tail -n 220 └ Jun 09 16:44:11 bing521 /usr/libexec/gdm-x-session[1633]: (WW) warning, (EE) error, (NI) not implemented, (??) unknown. … +96 lines (ctrl + t to view transcript) Jun 10 09:07:57 bing521 /usr/libexec/gdm-x-session[519779]: (–) NVIDIA(GPU-0): CXD Forced1080p (CRT-0): 400.0 MHz maximum pixel clock ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 现在关键结果已经出来了:xrandr 显示当前已经是 1920x1080,Xorg 日志也确认读取了伪 EDID 和强制配置。还有一 个自动回滚计时器挂着,我先把它取消,避免它把已经成功的配置删掉。 • Ran printf ‘%s\n’ ‘555999’ | sudo -S touch /run/codex-1080p-confirmed && printf ‘%s\n’ ‘555999’ | sudo │ -S systemctl stop codex-1080p-rollback.timer codex-1080p-rollback.service 2>/tmp/ │ codex-stop-final.err || true │ … +3 lines └ [sudo] password for bing521: NEXT LEFT LAST PASSED UNIT ACTIVATES 0 timers listed. • Ran ls -l /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf /etc/X11/edid/forced-1080p-edid.bin 2>/dev/ │ null || true └ -rw-r–r-- 1 root root 128 Jun 10 09:04 /etc/X11/edid/forced-1080p-edid.bin -rw-r–r-- 1 root root 1345 Jun 10 09:04 /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf • Ran xrandr --query | sed -n ‘1,18p’ └ Screen 0: minimum 8 x 8, current 1920 x 1080, maximum 16384 x 16384 DVI-I-0 connected primary 1920x1080+0+0 (normal left inverted right x axis y axis) 810mm x 460mm … +14 lines (ctrl + t to view transcript) 1440x810 60.00 59.97 1400x1050 74.76 59.98 • Ran cat /run/codex-1080p-revert.log 2>/dev/null || true └ (no output) • Updated Plan └ 重新安装已校验的 1080p EDID 和 NVIDIA Xorg 配置 重启图形服务让配置生效 用 xrandr 和 Xorg 日志验证 1920x1080 是否可用 如果失败,改用完整 xorg.conf 或内核 EDID override 方案 ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 已经好了。当前系统已经切到并支持 1920x1080: current 1920 x 1080 DVI-I-0 connected primary 1920x1080 1920x1080 60.00*+ 我做的处理是: 生成了一个通过校验的 1080p EDID:/etc/X11/edid/forced-1080p-edid.bin 写入 NVIDIA/Xorg 配置:/etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf 重启了 GDM 让配置生效 取消了自动回滚计时器,避免它删除配置 Xorg 日志也确认已经识别为 CXD Forced1080p,并设置了 1920x1080_60.00。目标已完成,用时约 24 分钟。 ─ Worked for 2m 30s ────────────────────────────────────────────────────────────────────────────────────── › 系统重启不会又变回去吧? • 我快速确认一下持久化文件和回滚任务状态。只要配置文件还在 /etc/X11,重启后 Xorg 会继续读取它。 • Ran ls -l /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf /etc/X11/edid/forced-1080p-edid.bin 2>/dev/ │ null || true └ -rw-r–r-- 1 root root 128 Jun 10 09:04 /etc/X11/edid/forced-1080p-edid.bin -rw-r–r-- 1 root root 1345 Jun 10 09:04 /etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf • Ran xrandr --query | sed -n ‘1,8p’ └ Screen 0: minimum 8 x 8, current 1920 x 1080, maximum 16384 x 16384 DVI-I-0 connected primary 1920x1080+0+0 (normal left inverted right x axis y axis) 810mm x 460mm … +4 lines (ctrl + t to view transcript) 2560x1440 59.96 59.95 2048x1536 85.00 75.00 60.00 • Ran systemctl list-timers --all ‘codex-1080p*’ --no-pager 2>/dev/null || true └ NEXT LEFT LAST PASSED UNIT ACTIVATES 0 timers listed. ────────────────────────────────────────────────────────────────────────────────────────────────────────── • 不会。现在是持久化配置,不是临时 xrandr: 配置文件还在:/etc/X11/xorg.conf.d/10-nvidia-force-1080p.conf EDID 文件还在:/etc/X11/edid/forced-1080p-edid.bin 自动回滚计时器已取消:0 timers listed 当前仍是:1920x1080 60.00*+ 所以正常重启后还会按这个配置进入 1920x1080。如果以后换显示器或换线导致异常,删除上面两个文件再重启 GDM/ 系统即可恢复默认识别。 ─ Worked for 1m 09s ────────────────────────────────────────────────────────────────────────────────────── › Find and fix a bug in @filename gpt-5.5 xhigh · ~/桌面 Goal achieved (24m) 解决了我困扰已久的不能用1080P的问题,/goal是真的强大 a-zA-Z0-9_- ↩︎ a-zA-Z0-9_- ↩︎ 3 个帖子 - 2 位参与者 阅读完整话题
1.你大厂职级我知道,相当于互联网圈的清北复交 2.你蹲大厂拧螺丝太屈才了,去AI startup早财务自由了 3.别聊吃喝玩乐了,我更想听你讲你是怎么不被裁员的 4.以前觉得互联网男都很飘,直到遇见你才知道什么叫真正的技术 5.好喜欢听你讲大模型逻辑,一点都不无聊 6.我爸囤了一堆4090显卡,让我问你需要吗,在家放着落灰了 7.你是不是故意单着吊小姑娘?这条件在互联网圈不早被抢疯了啊 8.平价格子衫穿你身上都像硅谷定制的老钱风 9.我朋友做具身智能,想找个有大厂背景的技术合伙人 10.别出去过520了,我想在家等你跑完模型看迭代数据 13 个帖子 - 13 位参与者 阅读完整话题
Every startup founder, creator, and small business owner faces the same problem early on: you need a brand, but you don't have a design budget. Hiring an agency costs thousands. Freelancers are hit-or-miss. And doing it yourself in Canva usually ends with a logo that looks like everyone else's. The good news? You no longer need to choose between "expensive" and "unprofessional." AI-powered brand identity tools have made it possible to create a complete brand identity from scratch —logos, mascots, color palettes, typography, merchandise, and style guides—in under an hour. In this guide, we'll walk through what a real brand identity system looks like, why visual consistency matters across every channel, and how founders are using BrandMaker to build scalable brand systems without starting over every six months. What Is a Brand Identity System (And Why Startups Need One Early) A brand identity system is more than a logo. It's the structured DNA of how your company looks, feels, and communicates visually. Think of it as the rulebook that keeps your brand recognizable whether someone sees you on Instagram, a billboard, or a product box. A practical brand identity system includes: Logo variations (primary, secondary, icon-only, reversed) Color palette with hex codes and usage rules Typography hierarchy (headings, body, accents) Mascot or visual character (increasingly popular for DTC and creator brands) Branded assets for marketing and social media (templates, banners, stories) Merchandise-ready designs (tees, stickers, packaging) A living style guide that your whole team can follow Without this system, most startups end up with fragmented visuals. One designer makes your Twitter banner. Another designs your landing page. A third creates your pitch deck. Nothing matches. You look like three different companies. The fix: Build a structured brand DNA from day one. And yes, you can do it online without a design background. How to Create Consistent Brand Visuals Online Consistency is the single most underrated growth lever in branding. According to brand research, consistent presentation across all platforms can increase revenue by up to 23%. But consistency is hard when you're moving fast. Here's how to create consistent brand visuals online, even if you're a solo founder juggling product, marketing, and customer support: 1. Start With Brand Fundamentals, Not Fancy Design Before you open any design tool, lock in your brand fundamentals: Who is your audience? What emotion should your brand evoke? (Playful? Premium? Rebellious? Trustworthy?) What makes you different from competitors? These answers become the filter for every visual decision. Without them, even a beautiful logo will feel disconnected from your product. 2. Use an Online Tool for Structured Brand DNA Creation Manual brand building in Figma or Illustrator is powerful—but slow. If you need to ship your MVP this month, you need speed without sacrificing quality. An online tool for structured brand DNA creation lets you: Answer a few strategic questions about your brand Automatically generate a complete visual system Export assets in formats ready for web, print, and social media This is exactly what BrandMaker does. It's an AI-powered brand identity system designed for founders who need professional results without the agency price tag. 3. Maintain Visual Consistency Across Products and Platforms Once your core system exists, the real work is using it everywhere. That means: Every Instagram post uses your brand colors and fonts Your landing page matches your product UI Your email newsletter feels like it came from the same company as your app Your pitch deck, investor updates, and job postings all share one visual language A brand design system beyond logos and icons is what separates hobby projects from serious companies. It's also what makes hiring designers later much easier—they inherit a system, not a mess. Build a Scalable Brand System for Product Teams If you're building a product with a team, your brand system needs to scale. What works for a solo founder won't work when you have marketing, sales, support, and external agencies all creating assets. A scalable brand system for product teams needs these properties: Property Why It Matters Documented rules Anyone on the team can create on-brand content Flexible templates Not every asset should require a designer Version control Your brand evolves, but old assets don't get orphaned Asset library Logos, icons, illustrations, and photos in one place Automated generation Bulk-create social posts, ads, and merchandise without manual design Most startups don't hit this level of organization until they're 50+ people. But with modern tools, you can set up a scalable system on day one. Generate Logo, Mascot, and Merchandise Automatically One of the most powerful shifts in branding AI is the ability to generate logo, mascot, and merchandise automatically from a single brand foundation. Here's why this matters: Logos That Actually Fit Your Brand Generic logo makers spit out templates. AI brand systems generate logos based on your actual brand strategy—your audience, your positioning, your personality. The result feels intentional, not recycled. Mascots That Build Emotional Connection Some of the fastest-growing brands (Think Notion, Duolingo, Discord) use mascots to build emotional connection. A mascot gives your brand a face people remember. With AI, you can generate a unique mascot that matches your visual style—no illustrator required. Merchandise That Markets Itself Creator economy founders know: merchandise isn't just revenue, it's marketing. When someone wears your tee or slaps your sticker on their laptop, they're endorsing you to everyone around them. An automated system can generate print-ready designs for tees, hoodies, stickers, and packaging in seconds. BrandMaker combines all three into one workflow. You don't design a logo, then separately commission a mascot, then hire someone for merchandise. You build one brand identity, and all assets flow from it automatically. Brand Identity Software for Creators and Entrepreneurs Creators and entrepreneurs have unique branding needs. You're not a Fortune 500 company with a 6-month rebrand timeline. You need: Speed (launch this week, not next quarter) Flexibility (your brand will evolve as you find product-market fit) Affordability (every dollar goes to product or growth) Quality (you're competing with well-funded companies for attention) Brand identity software for creators and entrepreneurs should solve all four. Here's what to look for when evaluating tools: Must-haves: AI-generated logos with real customization (not just font swaps) Complete style guide generation Export in multiple formats (PNG, SVG, PDF) Social media template generation Merchandise design output Nice-to-haves: Mascot generation Animated logo exports Brand voice / copy suggestions Team collaboration features All-in-One Brand Toolkit for Founders: What to Expect An all-in-one brand toolkit for founders should take you from zero to launch-ready brand in a single session. Here's a realistic timeline: Phase Time What Happens Brand Discovery 10 min Answer questions about audience, positioning, and personality Visual Generation 5 min AI generates logo, color palette, typography, and mascot options Selection & Refinement 15 min Pick your favorites, tweak colors, adjust layouts Asset Creation 10 min Generate social templates, merchandise designs, and marketing assets Export & Documentation 5 min Download everything + receive a living style guide Total ~45 min A complete brand system, ready to use Compare that to the traditional agency process: discovery call → proposal → contract → design rounds → revisions → final delivery. Timeline: 4–8 weeks. Cost: $5,000–$50,000. For most startups and creators, the speed and cost advantage of AI brand tools is a no-brainer. Grow Your Brand Identity Without Starting Over The biggest branding mistake startups make? Treating their brand as a one-time project instead of a living system. Your brand will evolve. You'll pivot. You'll launch new products. You'll enter new markets. You'll raise funding and need to look more premium. You'll go viral and need to capitalize on attention fast. A brand system that adapts as your company evolves is essential. That means: Your core identity should be flexible enough to stretch You should be able to generate new assets without redesigning from scratch Your style guide should be easy to update and share Your asset library should grow with you With a proper system in place, a rebrand becomes a refinement, not a restart. You're not throwing away six months of work. You're evolving what already exists. Design Professional Brand Materials Quickly: Real Use Cases Let's look at how real founders use automated brand tools in practice: Use Case 1: Pre-launch Startup Founder: Solo SaaS founder, no design skills, launching on Product Hunt next week. Need: Logo, landing page visuals, social media kit, and investor pitch deck—all on brand. Solution: Generate complete brand system in one afternoon. Export assets in every format needed. Launch with professional visuals that match the product quality. Use Case 2: Creator Building a Personal Brand Creator: YouTuber/podcaster with 50K followers, launching a paid community. Need: Logo, mascot, membership card design, merchandise line, and social templates. Solution: Create a brand identity that feels personal but scalable. Generate merchandise designs ready for print-on-demand. Build social templates that make every post instantly recognizable. Use Case 3: E-commerce Brand Founder: DTC product company preparing for a seasonal campaign. Need: Updated packaging, email banners, Instagram story templates, and influencer kit. Solution: Generate campaign-specific assets while maintaining core brand consistency. Produce 50+ social assets in an hour instead of a week. Turn Brand Fundamentals Into Usable Design Assets Strategy without execution is useless. The gap between "we know our brand" and "we have the assets" is where most startups get stuck. A consistent brand design workflow online bridges that gap. Here's the ideal flow: Define your brand fundamentals (audience, positioning, personality) Generate your visual system based on those fundamentals Create templates and reusable assets Apply the system across every channel Iterate as your brand evolves Tools like BrandMaker are built for this exact workflow. You don't need to export a logo and then figure out the rest yourself. The entire system generates together—logo, colors, fonts, mascot, merchandise, social templates, and a shareable style guide. Automated Brand Asset Generation With Style Guide A style guide is the backbone of brand consistency. Without one, your team will make inconsistent decisions every day. A modern AI-generated style guide should include: Logo usage rules (minimum size, clear space, incorrect uses) Color system (primary, secondary, neutral, functional colors with exact values) Typography rules (font families, sizes, weights, line heights for every use case) Mascot/illustration guidelines (when to use, how to style, what not to do) Photography direction (mood, lighting, subject matter) Voice and tone (if the tool supports it) Template library (pre-built designs for common needs) The best part? With automated brand asset generation with style guide , your style guide updates as your brand evolves. No more stale PDFs that nobody references. Build Cohesive Brand Visuals for Multiple Channels Your brand lives in more places than ever before. Here's a typical channel mix for a modern startup: Website and landing pages Web app / mobile app UI Social media (Instagram, Twitter/X, LinkedIn, TikTok) Email marketing Paid advertising (Meta, Google, LinkedIn) Content marketing (blog, YouTube, podcast) Sales materials (pitch decks, one-pagers, proposals) Physical touchpoints (packaging, merchandise, event booths) Internal communications (presentations, reports, job postings) Each channel has different constraints. Instagram needs square and story formats. LinkedIn needs professional banners. Your app needs icons and UI components. Packaging needs print-ready files. Building cohesive brand visuals for multiple channels only works when you have a system, not a collection of files. That's the difference between a brand identity system and a logo file. FAQ: Brand Identity Systems for Startups and Creators Can I really create a professional brand without a designer? Yes. AI brand tools have reached a level where the output rivals mid-tier freelance work for most startup needs. You'll still want a human designer for complex challenges or when you're Series B+, but for launch and early growth, AI-generated brands are more than sufficient. How is this different from Canva or a logo maker? Logo makers give you a logo. Canva gives you templates. A brand identity system gives you a complete, interconnected visual ecosystem—where every asset (logo, mascot, social post, merchandise) shares the same DNA and follows the same rules. Will my brand look like everyone else's? Not if you use the right tool. The key is starting with your brand strategy (audience, positioning, personality) and letting AI generate from there—not picking from a template library. The best AI brand tools generate unique outputs based on your inputs. Can I update my brand later? Absolutely. The best approach is to build a flexible foundation and evolve it. With a system-based approach, updating colors, fonts, or even your logo doesn't require starting from scratch. You regenerate assets from the updated system. What if I need to hire a designer later? Having a brand system actually makes hiring designers easier. You hand them a style guide and asset library instead of a chaotic Dropbox folder. They can focus on high-value work (campaign concepts, complex illustrations) instead of recreating your basic identity. Ready to Build Your Brand? You don't need a design degree. You don't need a $10,000 agency budget. You don't need to spend six weeks going back and forth with freelancers. What you need is a brand identity system that takes your strategy and turns it into a complete visual language—automatically. BrandMaker is an AI-powered brand identity system built for founders, creators, and entrepreneurs who need professional results fast. Generate your logo, mascot, color palette, typography, merchandise, social templates, and a living style guide—all in one place, all consistent with each other. Build your brand today. Not next month. 👉 Get started with BrandMaker
公司此前申请了 Kiro startup 项目 获得了一些初创积分 之前也尝试过 claude code 官方渠道 始终无法稳定使用 以为 Kiro 这条线好歹可以撑一撑 没想到今天也彻底没法用了 现在付费 Kiro 账号也只能使用些老旧的开源模型了 之前企业账户绑定的是香港的卡 不知道换成其他地区的卡是不是就能使用了 还是说一旦账户被判定为国区账户 换绑卡也没用 有相同经历的佬麻烦给点建议 另外 现在还有渠道能申请海外实体卡吗 貌似该封的好像都封的差不多了 目前也确实没啥好办法了 本来就是小团队 资金不足 现在 Kiro 这边积分浪费着 还得再掏钱寻求其他渠道 多少是有点可惜啊 1 个帖子 - 1 位参与者 阅读完整话题
MIT Technology Review This startup’s new mechanistic interpretability tool lets you debug LLMs Goodfire wants to make training AI models more like good old-fashioned software engineering. [!quote]+ 总部位于旧金山的初创公司 Goodfire 刚刚发布了一款名为 Silico 的新工具,它可以让研究人员和工程师窥视人工智能模型内部,并在训练过程中调整其参数–即决定模型行为的设置。这可以让模型制造商对这项技术的构建方式进行更精细的控制,这在以前是不可能实现的。 Goodfire 声称,Silico 是同类产品中第一个现成的工具,可以帮助开发人员调试开发过程的各个阶段,从建立数据集到训练模型。 该公司表示,它的使命是让建立人工智能模型不再像炼金术,而更像一门科学。当然,像 ChatGPT 和 Gemini 这样的 LLM 可以做出令人惊叹的事情。但是,没有人确切知道它们是如何工作的,也没有人知道它们为什么会工作,这就很难修复它们的缺陷或阻止不需要的行为。 Goodfire是少数几家公司之一,其中包括业界领先的Anthropic、OpenAI和谷歌DeepMind,它们都在开创一种被称为机械可解释性的技术,旨在通过绘制神经元及其之间的通路,了解人工智能模型在执行任务时的内部情况。(麻省理工科技评论》将机械可解释性选为 2026 年十大突破性技术之一)。 1 个帖子 - 1 位参与者 阅读完整话题
如果在启动 ClaudeCode 有以下 Hook 报错 SessionStart:startup hook error Failed with non-blocking status code: No stderr output 和这个 GPT-5.4经常性调用using-superpowers解决方法 也有一定关系 原因 如果使用的 Claude 安装的插件版本 Superpowers,会自带 SessionStart hook,在启动的时候会注入,但是注入的点为 startup|resume|clear|compact ,这会导致新启动的 ClaudeCode 会触发 non-blocking 错误,这也是为什么在使用 GPT 模型。尤其是 ClaudeCode 中,会优先调用 Superpowers 的原因,因为会在开头注入 using-superpowers 技能 “你必须在回复前检查是否有适用的技能” 技能优先级排序规则 各种 “不要跳过技能检查” 的红旗清单 修复 将这个 Json 文件中,改为如下这样即可(每个人版本可能不一样) ~/.claude/plugins/cache/superpowers-marketplace/superpowers/4.0.3/hooks/hooks.json { "hooks": { "SessionStart": [ ] } } 注意!插件更新后该文件会被还原,需重新修改。 1 个帖子 - 1 位参与者 阅读完整话题
有朋友申请过 Google for Startups Cloud Program 吗? 2 个帖子 - 2 位参与者 阅读完整话题