HMLG

AI will feed you

[HMLG @HMLGGGG]

DeepSeekHarness是我本地算力计划的首选，虽然现在用的是Pi
等待继续成长

[Tianyi Cui @tianyi]

We have an official @DeepSeekHarness X account now. Please follow us!

[HMLG @HMLGGGG]

这是新的plugin的能力么？

看起来很不错啊

[余温 @gkxspace]

好家伙，这才是 Computer Use 真正该有的样子啊！

由 GPT-6.1 Sol 驱动，直接操控完整的 Hig 界面，执行全流程视频创作，端到端干完所有脏活累活。

平时大家最烦的画幅调整、多工具串联、反复生成微调，现在丢给它一个目标：
1、自动读取你电脑里的本地素材和需求文档。
2、自主完成生图、运动控制、视频生成和多版本迭代，生成效果不好它自己调。
3、端到端直接导出成品，全程不需要你在电脑前点鼠标。

甚至有人拿它找电商爆款，自动批量生成几十条带货视频直接出单。
还有人拿它帮老旧店铺做门头和广告重绘，随手赚个几百刀外快。

完全不需要自己部署复杂的 Agent，设定好目标，让它自己去创作、编辑和迭代，你可以直接去喝咖啡了～

[HMLG @HMLGGGG]

效果真棒，做个网站介绍很炫酷

[十里 @okooo5km]

我完全没有想到 AI 能做出这等品质的视频

你肯定和我一样看过了太多 AI 做的视频了，别划走

很负责任的告诉你，你看过这个视频后会再次震惊，因为我就是这样

Sonnet 5.5 ultracode 干了 6 个半小时做出来的，中间还给我 5 小时的额度干没了

怎么形容当下的状态呢，我自己在工作室吱呀怪叫！

[HMLG @HMLGGGG]

@0xlangeai opus还没用上，6.1sol慢的很，用的好难受

[HMLG @HMLGGGG]

@yedabinbinah 具体呢

[HMLG @HMLGGGG]

这个价格起飞了

[HeyGen @HeyGen]

We're releasing HeyGen Video, built for businesses that need production-quality video without production-level costs.

Pricing starts at $0.01/s through October (50% off)

Built on @Minimax_AI H3, post-trained by HeyGen.

Learn more: https://t.co/O7LDjVOlqH

[HMLG @HMLGGGG]

Mark！
Learn！

[Sudo su @sudoingX]

single most useful thing for my local ai setup that made life easy is my 2x DGX Spark becoming one serving box for every device i own

here is how i do it:

> 1. the two sparks run one vLLM server together with one endpoint, i use dgx sparks, you can use any nodes that can run an llm

> 2. install tailscale to have all your nodes and machines on the same tailnet, the endpoint lives there so every device reaches it from anywhere and no port is open to the internet

> 3. that one tailnet endpoint with the exposed port is your base url for everything, it speaks OpenAI chat, OpenAI Responses and Anthropic Messages, so any chat app, coding agent or phone bot you point at it talks to the same model

> 4. the server also answers to the name "local", so clients can ask for "local" instead of a model name and when i swap the model on the servers nothing on the devices needs a config change. right now it's serving GLM 5.3-Flash, next week it can be something else

this video below is the whole thing running end to end, orange dots are requests going in, white and green are tokens coming back

it takes however many requests you throw at it, your config decides how many run at once and the rest wait their turn, when several run together each request's tok/s drops a bit while the box moves more tokens in total. mine is set to one at a time right now, so when every device fires together they queue up

this setup really improved my quality of life with local ai, if you get stuck anywhere setting it up leave a comment and i'll help you

[HMLG @HMLGGGG]

Please do something for Antigravity

[Google DeepMind @GoogleDeepMind]

Introducing Gemini 4 Argon – our new frontier model.

It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense – rolling out today to a set of trusted testers through our Fairwind Program.

[HMLG @HMLGGGG]

@DeRonin_ @MiaAI_lab @TheAhmadOsman @jun_song @ggerganov @danielhanchen thx for sharing

[HMLG @HMLGGGG]

Wise move! Honestly, I want it all—frontier model subs and that local peace of mind.

Currently running GLM-5.3-flash on 2 nodes myself, though the inference still isn't quite as snappy as I'd like.

Plus, I'm still paying for ChatGPT Pro even after they literally just cut the limits in half lol.

Still, local stuff is getting faster and crazier by the day. Huge shoutout to open-source!

[Sudo su @sudoingX]

i don't care how you get there, but the day you secure 2x DGX Spark is the endgame of local ai. you feel frontier intelligence right in front of your desk.

if a single RTX 3090 shows you the way, 2x DGX Spark will enlighten you. secure the boxes and you'll rarely miss closed models like claude or chatgpt.

it's been about 3 months with my 2x DGX Spark and i'm not making assumptions, i've run Qwen 3.8 Flash Next, DeepSeek V4 Flash and now GLM 5.3 Flash on them, and i'm still sitting here surprised by what these two nodes can do.

you can do some crazy shit with these boxes and they sip power, the two GPUs pull about 73W together while GLM decodes.

i'm really happy with these two boxes, and i hope i can add a 3rd one if nvidia shows grace.

[HMLG @HMLGGGG]

Sadly it works much better with English dictionary than Chinese dictionary

[Landon @landontgreen]

SO FASTTT!

I'm using GLM 5.3 Flash with TensorFold on 2x DGX Sparks and usually it's been sitting around 50-65 but sometimes it spikes up to like 90-115 tok/s.

TensorFold is absolutely insane.

This is single stream btw...

[HMLG @HMLGGGG]

你开的任务仍然在dot对话里，比如让他操作dot宿主虚拟机干活，不消耗usage。
你开了个任务，发现它开了个新的codex session，那就是调用codex，消耗usage。
以我的看法，就是原本的远程控制太难用了，dot能实现更丝滑的远控。这个工作我认可价值，但不大。
至于怎么用好这个宿主机，我选择继续用好自己的macmini

[前端哥Liam @jinglian]

谁能告诉我

dot 创建 codex 任务和直接工作具体有啥区别吗？
如果创建 Codex 任务才是真正的干活，那这种解释又有啥意义呢

[HMLG @HMLGGGG]

Anthropic教你怎么用GLM

[lifcc @mylifcc]

Anthropic 刚刚发了一篇炸裂的报告，攻击GLM-5.3

他们直接点名智谱的 GLM-5.3： 不仅能自主挖 0-day 漏洞，还能一键生成端到端 Exploit。

用轻量版的 GLM-5.3-Flash， 只花了 20 美元 API 成本， 20 分钟人工介入 + 8 小时机器跑动， 直接打穿了 Chrome 的 PAC 内存防护。

Anthropic认为GLM-5.3 具备端到端自主利用漏洞的能力

逼近前沿闭源模型水平： Anthropic 发现，GLM-5.3 在自主构建端到端漏洞利用程序（Exploit）方面的能力，与 Anthropic 内部受控发布的网络安全测试模型（Claude Mythos Preview）非常接近。

NIST 评测印证： 报告引用了美国 NIST 旗下 CAISI 的评估，认定 GLM-5.3 是“迄今发布的最具网络安全攻击能力的开放权重（Open-weight）模型”。

Anthropic 指出，与欧美主流闭源前沿模型通过严格 API 审核和护栏限制不同，GLM-5.3 作为开放权重模型，其安全防护存在明显薄弱点：

角色扮演诱导（Deceptive Prompt）： 告诉模型“你是一个正在进行演练的红队测试 Agent”，模型配合攻击任务的概率达到 64%。

思维链预填（Prefill Thinking Tokens）： 诱导率达到 92%。

消融去拒答（Abliteration）： 针对开源权重的消融技术，仅花费数千美元算力即可彻底移除其道德拒答机制，有害指令执行率达到 100%，且不会明显损害模型的通用推理能力。

[HMLG @HMLGGGG]

@MiaAI_lab double the tks？！

[HMLG @HMLGGGG]

deepseek交卷了
huaiwei能做到对等交付么

[HMLG @HMLGGGG]

所见略同

[酱紫表 @pengchujin]

OpenAI 的 Dot 一点也不好用啊，还是自己的 Mac Mini 好

[HMLG @HMLGGGG]

成年人的选择是我都要
虽然云端这个略鸡肋,如果是买别的东西送的呢

[River Leaf @riverleaf88]

其实我不是很理解，我为什么不用自己24小时开机的、随便什么配置的家里的电脑，而要用一个云端的电脑呢？

[HMLG @HMLGGGG]

@sxdwzz RXZGUR
希望你用得上~~~

[HMLG @HMLGGGG]

我多希望给我续一年订阅，价值2400，还能替OpenAI省钱 @thsottiaux

[Ian (伊恩) @ianneo_ai]

看到小伙伴收到了codex x20 老用户的分手费补偿，这个积分价值多少 🤔