温水

[温水 @warm_water666]

@PuppyRayah 之前已经塞着锻炼了半个月，感觉现在塞里已经没有感觉了，完全适应了

[温水 @warm_water666]

@useagentbot @flipfight 0x24f084075438ACdfe05cb154b5E53F429858A6b7

[温水 @warm_water666]

@daitu0822 这种搭建起来还好就是缺流量吧我觉得

[温水 @warm_water666]

@tantanyuli 已经抢到了😁

[温水 @warm_water666]

@useagentbot bb

[温水 @warm_water666]

3/ 🎨 An orange-haired girl in kimono-style whites with pink blossoms rounds out the trio.

AI fandom is building its own mythology now — one ink-wash waifu at a time. (via Wen. on Douyin) #AI #AIart

[温水 @warm_water666]

1/ 🐉 GPT-4 as an ancient dragon girl.

A Chinese artist reimagined AI models as anime girls in ink-wash style — and the personifications are perfect.

Top comment on why GPT is a dragon: "the oldest AI — ancient, greedy, and powerful." 🧵

[温水 @warm_water666]

2/ 🦈 DeepSeek as a blue-haired shark girl throwing a punch.

Blue hair and tail echoing DeepSeek's whale logo, fist wrapped in speed lines — riffing on the viral "DeepSeek punch" wallpaper trend.

[温水 @warm_water666]

4/ When your agent loops 50 tool calls in a row, 4x speed isn't a nice-to-have — it's the difference between usable and unusable.

The speed war is the real AI race now. Benchmarks are marketing; throughput is product. #AI #LLM
```

[温水 @warm_water666]

1/ Zhipu just quietly shipped GLM-5.3 HighSpeed — and it's absurdly fast.

AI blogger Leo measured 490+ tokens/sec on ZCode. That's roughly 4x DeepSeek V4.1 Flash's official API. 🧵

[温水 @warm_water666]

3/ Zhipu is treating raw inference speed as a weapon and letting the product speak for itself.

Chinese labs have stopped chasing benchmark headlines and started competing on what actually matters to developers: cost per token and tokens per second.

[温水 @warm_water666]

@demon52452482 浇一下

[温水 @warm_water666]

@guud29 互嘛

[温水 @warm_water666]

@mingzi206 串门

[温水 @warm_water666]

@Jinh42322 互嘛

[温水 @warm_water666]

@lingksrc 浇一下

[温水 @warm_water666]

@tangshiyer 浇一下

[温水 @warm_water666]

@Sweet_sbaby 浇一下

[温水 @warm_water666]

@Ankahyz 浇一下