David's random thoughts
Open in Telegram
个人想法合集，主要同步来自Twitter (𝕏)、博客等账号发布的内容。 注：本频道并非纯粹包含技术相关内容（虽然以它们为主），本人不会刻意回避政治观点表达。可能包含一些直言不讳的主观评价，慎入。 个人博客：https://blog.hjc.im
Show more1 935
Subscribers
-524 hours
-37 days
-1130 days
Data loading in progress...
Similar Channels
Tags Cloud
Incoming and Outgoing Mentions
---
---
---
---
---
---
Attracting Subscribers
August '26
August '26
+73
in 6 channels
July '26
+95
in 11 channels
Get PRO
June '26
+64
in 1 channels
Get PRO
May '26
+35
in 1 channels
Get PRO
April '26
+72
in 7 channels
Get PRO
March '26
+14
in 3 channels
Get PRO
February '26
+56
in 5 channels
Get PRO
January '26
+43
in 6 channels
Get PRO
December '25
+27
in 6 channels
Get PRO
November '25
+22
in 0 channels
Get PRO
October '25
+13
in 0 channels
Get PRO
September '25
+27
in 2 channels
Get PRO
August '25
+25
in 2 channels
Get PRO
July '25
+10
in 0 channels
Get PRO
June '25
+41
in 7 channels
Get PRO
May '25
+40
in 2 channels
Get PRO
April '25
+42
in 3 channels
Get PRO
March '25
+55
in 10 channels
Get PRO
February '25
+73
in 5 channels
Get PRO
January '25
+54
in 13 channels
Get PRO
December '24
+113
in 9 channels
Get PRO
November '24
+113
in 8 channels
Get PRO
October '24
+222
in 16 channels
Get PRO
September '24
+92
in 6 channels
Get PRO
August '24
+64
in 6 channels
Get PRO
July '24
+517
in 10 channels
Get PRO
June '240
in 12 channels
Get PRO
May '240
in 5 channels
Get PRO
April '240
in 4 channels
Get PRO
March '240
in 5 channels
Get PRO
February '24
+6
in 15 channels
Get PRO
January '24
+46
in 2 channels
Get PRO
December '23
+36
in 1 channels
Get PRO
November '23
+24
in 3 channels
Get PRO
October '23
+263
in 1 channels
| Date | Subscriber Growth | Mentions | Channels | |
| 30 August | 0 | |||
| 29 August | +4 | |||
| 28 August | +2 | |||
| 27 August | +2 | |||
| 26 August | +2 | |||
| 25 August | 0 | |||
| 24 August | +2 | |||
| 23 August | 0 | |||
| 22 August | +2 | |||
| 21 August | +5 | |||
| 20 August | +13 | |||
| 19 August | +3 | |||
| 18 August | +4 | |||
| 17 August | +2 | |||
| 16 August | 0 | |||
| 15 August | +1 | |||
| 14 August | +2 | |||
| 13 August | +1 | |||
| 12 August | +2 | |||
| 11 August | +4 | |||
| 10 August | 0 | |||
| 09 August | +3 | |||
| 08 August | +2 | |||
| 07 August | +2 | |||
| 06 August | +2 | |||
| 05 August | +2 | |||
| 04 August | +2 | |||
| 03 August | +3 | |||
| 02 August | +4 | |||
| 01 August | +2 |
Channel Posts
才发现这里有一家同时从Web和移动端两条路线大幅度扩大无效内存需求的万恶之源公司居然开始要求开发者节约内存了🤮
https://android-developers.googleblog.com/2026/08/app-quality-memory-optimization-secure-onboarding.html
| 2 | 前有27B战284B，后有320B战2.8T
刷起分来演都不演的呗 | 687 |
| 3 | 更新Intel 386H （DDR5-5600 SODIMM）的全套测试成绩。感觉还是首发的358H强……
https://benchview.hjc.im/#dataset=data%2FSPECrate2026_int_base.csv&base=Intel+Core+Ultra+9+386H+P-core+%28DDR5-5600B+64G%29 | 965 |
| 4 |  Intel 386（不 | 1 005 |
| 5 |  自从玩过wasm之后我再也不嘲笑5G是无效基建了（图为brotli压缩后的web app） | 1 085 |
| 6 | 3年过去了，Twitter/X的spam大概是让这平台真的已经完蛋了吧。当然另一方面Telegram也几乎变成Spam-as-a-Service了，好在有llm admin bot帮我自动化清理评论区还算勉强能用😅
https://x.com/hjc4869/status/1675401146607738882 | 1 920 |
| 7 | 😅 GPT再不改改这没活硬整的臭毛病就要比我古法编程还慢了 | 1 546 |
| 8 | 在 AvaloniaUI 软件中渲染 HDR 视频内容 https://blog.hjc.im/avaloniaui-hdr.html | 1 445 |
| 9 | 多屏幕配置下的Windows总是能精准地让新窗口被创建在我最不想它出现的那个屏幕上 | 1 508 |
| 10 |  每次看到这个都不自觉的脑补成 .DS_Store | 1 961 |
| 11 | 美西方恶意抢夺科技话语权😡 /s
https://github.com/anomalyco/opencode/pull/41532 | 3 062 |
| 12 | AI分析师们一个个都引用MS财报里卖出多少M365 copilot订阅来洋洋洒洒的分析AI的前景和盈利能力，可是我看到的用公司订的copilot的员工怎么没一个是不对这货骂娘的😂 | 2 286 |
| 13 |  3个月电费账单A$393，这就是为什么我一直想把家用server从9955HX换成PTL…… | 2 095 |
| 14 | 本来现在gcc在生态上就已经被llvm远远甩在身后，再拒绝AI生成代码会怎么样我不好说。。。
https://www.phoronix.com/news/GCC-Declining-AI-Contributions | 2 148 |
| 15 | K3本质上还是个闭源模型，所谓开源主要是一种营销手段。
虽然我这两天吐槽了A\和OpenAI等一堆商业公司，但是现在很多人对开源模型有奇怪的期待，也是时候该泼下冷水了。开源并不能省钱，你算出来便宜只是在白嫖别人的训练成本，长期看这个钱你还是得以某种形式付出去。
当然我100%支持模型厂商收钱。
https://x.com/cherylnatsu/status/2081996644476965119 | 3 591 |
| 16 | 从未主张禁止开源AI模型，但是主张禁止我以外的模型
https://x.com/AnthropicAI/status/2081864750296658008 | 3 948 |
| 17 | 可是他自家的open model就是一坨巨大的狗屎，说这么多鬼话到头来还是想白嫖中国公司融资的钱然后倾销到美国去跟OpenAI/Anthropic恶性竞争罢了（
https://x.com/hjc4869/status/2080658505732858223
现在看到那几个美国公司谈开源模型就想笑。
两三年时间，要么是造了几吨Nemotron那种下载嫌浪费流量的不可回收垃圾，要么发一次gpt oss就没了下文，要么弄phi那种改几个字就跑分骤降的过拟合刷分神器。没哪家拿出真正SOTA方案给客户本地部署。
现在倒是想起来还有开源模型这回事了，怎么这么巧呢？ | 2 467 |
| 18 | 那你的gpt oss呢😅
https://x.com/sama/status/2080683363174945065 | 1 561 |
| 19 | 所以我一般都会让AI先写repro人工审一下然后再让它去修
https://x.com/xicilion/status/2080552376230887541 | 648 |
| 20 | 我用最直接、最真相、最不绕弯的方式来告诉你，如果一个跑分软件3年大改3次评分标准，那不如大家都别测试了，直接找厂商搞竞价排名。 | 1 579 |
