Zartbot

[Zartbot @zartbotF]

A short talk on Recursive Self-Improvement(RSI)

https://t.co/ywmZ9CfTkl

[Zartbot @zartbotF]

@holegots 哈哈哈, 你又锤...

[Zartbot @zartbotF]

@Kimi_Moonshot Excellent !

[Zartbot @zartbotF]

I just wrote an analytical report on Linear Attention and Sparse Attention, thoroughly discussing their respective strengths and weaknesses from both mathematical and engineering perspectives.

https://t.co/lAJlSSFXrM

[Zartbot @zartbotF]

Yesterday, NVIDIA published a blog post about the Rubin architecture, and I conducted some detailed analysis based on the preview version of the PTX 9.4 Instruction Set guide.

https://t.co/Ap7KPoV7nV

[Zartbot @zartbotF]

@holegots 还有我们美国豆包 Gemini

[Zartbot @zartbotF]

@xiangyuli 是，好用

[Zartbot @zartbotF]

it's true :)

[Venkatesh @Venkydotdev]

for real 😂

[Coin Bureau @coinbureau]

BREAKING: Alibaba says its new AI model is "second only" to Anthropic’s Fable 5.

The Chinese tech giant says its Qwen3.8 Max has 2.4T parameters, comparable to leading frontier AI models and second only to Fable 5.

The ranking is Alibaba’s own claim and has not yet been independently verified.

[Zartbot @zartbotF]

There's a homework assignment in the Anthropic interview about architectural optimization with VLIW+SIMD. It's a really interesting problem.

I spent a few hours on it and have already gotten it down to 987 cycles.

https://t.co/Zt86eNXheb

[Zartbot @zartbotF]

@ewind_dev 这也叫工程？大作业都算不上

[Zartbot @zartbotF]

@LeotheHorseman @qiqileyuan :)

[Zartbot @zartbotF]

@FallMonkey @jukan05 ？why?

[Jukan @jukan05]

Recommended weekend read.

https://t.co/i6WrmGgUok

[Zartbot @zartbotF]

@jukan05 maybe I need Fable 5 translate into english and publish in somewhere :)

[Zartbot @zartbotF]

It's a super complex task. My plan^3 is so convoluted that even I'm getting dizzy trying to wrap my head around it. There's just no way I can describe it to the model in plain language. https://t.co/vqXzIOK2rM

[Zartbot @zartbotF]

Recently, I used an AI agent to develop a library for scatter host-to-device memory copying—without writing a single line of code myself, The final implementation achieved several times the performance of cudaMemcpy.https://t.co/KYgdEFUeH7

[Zartbot @zartbotF]

@jukan05 :)

[Zartbot @zartbotF]

Recently, I wrote an article on processor architecture and interconnects in the Agent era.
https://t.co/Jz4pdADt6N