赛博吴同学

♨️公众号/视频号：@赛博吴同学

[赛博吴同学 @jax_wu]

@gptchan404 终于改成 go plus pro max 了。以前那套名字谁记得住

[赛博吴同学 @jax_wu]

@Layton_Gott 匿名藏了八个月没人认。红绿蓝突然一起冒出来，这下更说不清了

[赛博吴同学 @jax_wu]

@loveAIwang K3.1 闪了一下就没了。用出包浆的 K3 还得再等等啊

[赛博吴同学 @jax_wu]

@CainArch 解码快一倍，agent写内核人都审不过来了。这速度有点猛

[赛博吴同学 @jax_wu]

NVIDIA 没换芯片，硬是把 2.8T 的 Kimi K3 撸到 423 tok/s。靠的是 CentML 那人写了个一万二千行的 decode megakernel，塞进 vLLM。

同样硬件快 1.68 倍。挺离谱。

大家天天盯模型参数，这回赢的是水管工。注意力、MoE、通信揉进一颗核里，软件把硬件榨干了。

下一刀说是 DeepSeek V4.1。模型卷完了，推理栈接着卷啊

[SemiAnalysis @SemiAnalysis_]

BREAKING NEWS🚨: NVIDIA has just achieved 423 tok/s/user 🚀 on the 2.8T Parameter Kimi K3 model via a new upstream vLLM optimization. 1.68x Faster on the Same Hardware!

This was done by NVIDIA CentML's Benjamin, who wrote NVIDIA's new decode-specialized kernels in Cute DSL via aggressive Megakernel fusion. This is a 12,000 LoC Megakernel and is the future of advanced AI-assisted kernel generation. NVIDIA is already working on DeepSeekv4.1 and other models for vLLM too.

Someone should give this NVIDIA 10x engineer a raise for this great new major feature!

[赛博吴同学 @jax_wu]

@HenyJone 不改权重，让 Agent 自己改工作台。这思路挺野

[赛博吴同学 @jax_wu]

@AndrewCurran_ @SmokeAwayyy 三个都还没见过，他就先押蓝。挺急

[赛博吴同学 @jax_wu]

@Staraishow 又一家把Key弄丢了。一晚上刷成负数，服了

[赛博吴同学 @jax_wu]

@yaohongan2025 现成机器人直接捡着用。自己从零搭的那套先放放吧

[赛博吴同学 @jax_wu]

OpenAI API 里突然冒出三个没公布的模型：aurora 红绿蓝。

圈里猜是编程专用或者高速推理。以前 OpenRouter 那个 aurora-alpha 后来对上 Codex-Spark，这次也大概率同一套路。

没定价、没文档、还调不了。alpha 三个字摆那儿呢。

挺离谱。现在发模型都不用发新闻稿了，先让监控号扫到 ID，社区自己吵半周。

真能用再说。先别往生产里塞

[Lisan al Gaib @scaling01]

these just appeared in the OpenAI API:

aurora-blue-alpha
aurora-green-alpha
aurora-red-alpha

there was an openrouter model called aurora-alpha earlier this year which was likely GPT-5.3-Codex-Spark

so these could also be specialized codex coding models, or high-speed inference versions of GPT-6 models

[赛博吴同学 @jax_wu]

@AChaiNotes 搜不到也不收钱。这规矩挺厚道

[赛博吴同学 @jax_wu]

无限潜力🤙(⁎⁍̴̛ᴗ⁍̴̛⁎)可以多适配点开源模型吗？？最好是最新版本！

[SemiAnalysis @SemiAnalysis_]

BREAKING NEWS🚨: NVIDIA has just achieved 423 tok/s/user 🚀 on the 2.8T Parameter Kimi K3 model via a new upstream vLLM optimization. 1.68x Faster on the Same Hardware!

This was done by NVIDIA CentML's Benjamin, who wrote NVIDIA's new decode-specialized kernels in Cute DSL via aggressive Megakernel fusion. This is a 12,000 LoC Megakernel and is the future of advanced AI-assisted kernel generation. NVIDIA is already working on DeepSeekv4.1 and other models for vLLM too.

Someone should give this NVIDIA 10x engineer a raise for this great new major feature!

[赛博吴同学 @jax_wu]

@SemiAnalysis_ 无限潜力🤙(⁎⁍̴̛ᴗ⁍̴̛⁎)可以多适配点开源模型吗？？最好是最新版本！

[赛博吴同学 @jax_wu]

@TeslaZoa 应该设计到客厅天花板上🤩

[赛博吴同学 @jax_wu]

应该设计到客厅天花板上

[TeslaZoa @TeslaZoa]

Could this be the future of parking? 🚗✨

[赛博吴同学 @jax_wu]

@_FORAB 卖光家当跑了两千公里，结果人是AI编出来的。卧槽

[赛博吴同学 @jax_wu]

Anthropic连发两份内部报告：1334名员工预计2030年AI能做完自己75%的工作，普通人只敢选28%。公司里Agent工时已超人类6倍，工程师每周要读424小时的AI输出——比一周168小时还多一倍。更惊人的是，造前沿模型的活，AI已经干了51%，到2030年只剩5%留给人。

#赛博吴同学 #AI #Anthropic #AI取代人类 #大模型 #Claude #SI #AGI #RSI #程序员 #开发者

[Lisan al Gaib @scaling01]

Over half of surveyed Anthropic staff believe AI will be able to make Nobel Prize-worthy discoveries by 2027

[赛博吴同学 @jax_wu]

@scaling01 好期待 DeepSeek 5 的样子

[赛博吴同学 @jax_wu]

@imjustnewatai DeepSeek 5 跟着？🤗

[赛博吴同学 @jax_wu]

@Razer 算你牛