Dango233

Baking open AI system
Garnishing open weights

[virushuo @virushuo]

很多人不知道的历史：nvidia上海张江研发中心是在cuda 1.0发布之前就成立了。很大一部分cuda早期工作是在张江做的。中国不少大学也是最早开始尝试使用cuda的。整个gpu计算和ai发展大部分时间都是全球合作的产物，最后落得今天这样实在是幻灭。

[Dango233 @dango233max]

@op7418 我直接平了
被迫倒地

[Dango233 @dango233max]

亦有贡献，非常开心

[antirez @antirez]

After hte latest changes, DwarfStar can run DS4.1Flash in SSD streaming in a quite fast way. If it wasn't for the overthinking of course, that totally *did not* effect 4.0 Flash. It's a 4.1 flaw partially fixed in the latest update.

[Dango233 @dango233max]

@crystalwizard Mine started to talk defensively exactly in the GPT tone and triggered my PTSD

[Dango233 @dango233max]

目前没有足够证据断言 Grokbot 已经凉了，但现有观察与“正在变凉”的假设相符。

需要明确的是，“说话像 GPT”并不能推出“后端就是 GPT”。这是两个相关但不同的命题，不宜混为一谈。

现有证据能够支持的较窄结论是：我不需要两份 GPT 订阅。

我正在考虑退订 X 和 Grokbot。当然，这不应被解读为正式的退订决定，而是对我是否愿意继续为这坨东西付钱的阶段性重新评估。

[Alexi Gladstone @AlexiGlad]

Energy-Based Models can now scale better than Diffusion models.

Introducing TEM: Transformer Energy-based Models

TLDR:
• TEM outscales Diffusion without guidance
• First scalable contrastive Energy-Based Models
• Better generalization, test-time scaling, and OOD detection than Diffusion

🧵

[Dango233 @dango233max]

我宣布我有 “不把未完成写成已交付” 的PTSD，暂时隔离Astra/Sol一段时间

[Dango233 @dango233max]

Wait, what's going on with this second image!?

[DeepSeek @deepseek_ai]

🚀 DeepSeek-V4.1-Flash-1017 is here!

Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API.

🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.

1/4

[Dango233 @dango233max]

@FishAudio DRAMA

[Dango233 @dango233max]

@FishAudio Drama

[Dango233 @dango233max]

The answer is blowing in the wind.

[Tiezhen WANG @Xianbao_QIAN]

How many models are going to be released next week 😂

[Dango233 @dango233max]

@bot @bot get me my mail [email protected]

[Intelligent Internet @ii_posts]

A 30M-parameter model, on CPUs, inside Postgres

It matches a 0.6B dense embedding model on top-100 recall, to within 0.004, and finds more in the top 1,000

Meet Evoke. Open source today.

[Dango233 @dango233max]

@LeaskH 一秒笑喷的我成分也是很复杂

[Dango233 @dango233max]

Q: What would you do if you had unlimited tokens?
A:

[Dango233 @dango233max]

@LeaskH 自从Astra出现了一个特别2的Case:
任何东西只要被内置浏览器拦了一次，你不要想再把这个东西打开了，全部被判定成绕过。让他起server啊等等全不行，开发都不要做了

[Dango233 @dango233max]

@LeaskH @hfcorriez sol的tk/s也能赢哈哈哈哈

[Dango233 @dango233max]

@LeaskH @hfcorriez 6sol/luna虽然被骂的飞起，但是搞自动化是真的划算

[Dango233 @dango233max]

@oran_ge 设计和品味确实甩了Astra两条街，但是硬核工程现在感觉好像还是不如astra...

[Dango233 @dango233max]

Something doesn't look right here...