Norias

Interneticien lindois

[Norias @Norias_]

On peut faire des trucs sympas avec un tableau de l'artiste peintre Zigou (https://t.co/1RyVZ0T6f6) animé par une vidéo d'un poème de Fernando Pessoa lu par .@mariademedeiros :)

Thx @tg_bomze for the Face-Image-Motion-Model colab and @akai_katto for the dandere2x docker image !

[Linoy Tsaban @linoy_tsaban]

loving this new outpaint LoRA for Krea 2
small, great quality, super fast with Krea 2 Turbo

https://t.co/gcMoRil8Qt

[AI Breakfast @AiBreakfast]

Don’t get too excited, you’d need about 8x H100s to run Kimi 3 locally

[Arena.ai @arena]

Big news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5.

This is a 17-place jump from Kimi-k2.6 (#18 -> #1).

In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5.

The full model weights will be released by July 27.

Congrats to the @Kimi_Moonshot team on this major milestone!

[PhotoX86 @PhotoX86]

上次提到的krea2-identity-edit这个lora已经升级到了v1.1版本，似乎效果提升了一些。之前的工作流中使用的是comfyui-krea2edit节点，这个节点只能连接两个参考图，这次使用的是ComfyUI-Krea2-Ostris-Edit节点，它可以使用三个参考图，这样，人物、服装和场景三个项目就都可以有参考了。节点的链接地址https://t.co/6BPa4YXgvh。（提示词：图片1的人物穿着图片2的衣服站在图片3的场景中做出跳舞的姿势。）

尽管 Krea 2 本质上是 t2i 模型，krea2-identity-edit配合ComfyUI-Krea2-Ostris-Edit通过 Qwen3-VL 文本编码器融合图像和提示词，让 Krea 2 模型在生成时能“看见”并参考输入的参考图像，实现类似 i2i 或 edit 的效果。

[OpenClaw🦞 @openclaw]

🦞 #100000

100,000 issues + PRs in 222 days.

🛠️ built by volunteers
🌍 every timezone, every day
🧡 zero VC, one lobster

Number 100000 itself? A community bug report. We'll fix that one too.

Thank you for building this with us.

[memo akten @memoakten]

11 years old today.
#Deepdream is blowing my mind. And it's not the aesthetics.
https://t.co/aLKWN3iXkt
#inceptionism @zzznah @mtyka @ch402

[Peter H. Diamandis, MD @PeterDiamandis]

Alibaba allegedly ran 28.8 million fraudulent API exchanges across 25,000 fake accounts to steal Claude's intelligence. If confirmed, it's the largest AI model theft ever attempted. The same week, the White House restricted GPT 5.6 to 20 companies, OpenAI delayed its IPO, and Neuralink announced it may attempt brain-to-brain telepathy this year.

-- GPT 5.6 launches in three tiers: Sol, Terra, Luna all throttled by the White House.

-- Chinese proxy services offer Western frontier models at 90% discount. The trade: your reasoning traces get harvested for distillation.

-- OpenAI won't IPO below $1 trillion. Revenue is at $40–50B annually but they're burning $26B.

-- Elon's endgame for Neuralink is I/O layer for the singularity, allowing humans to couple with AI directly.

[Maziyar PANAHI @MaziyarPanahi]

Got GLM-5.2 running on my Mac Studio via llama.cpp, the reasoning behind all my medical agentic workflows.

It orchestrates a swarm of tiny on-device OpenMed experts: oncology, meds, labs.

No cloud, no rate limits, nobody can take it away.

AI must be owned, not rented.

[Kevin Lin @KevinQHLin]

thanks @_akhaliq for sharing our Data2Story!

🔮Turn the ‘Humanity's Last Exam’ dataset into a generative blog.

Explore more agent-generated stories here
https://t.co/TIIMWBamyU

[Z.ai @Zai_org]

GLM-5.2 leads GLM-5.1 by a wide margin across various domains, including coding, tool usage, reasoning, and general knowledge.

[Z.ai @Zai_org]

Introducing GLM-5.2: Frontier Intelligence, Open Weights

- Significant improvements in coding and agentic tasks
- Strong long-horizon capabilities with a 1M context window
- Two levels of reasoning effort: GLM-5.2 (max) pushes the limits, while GLM-5.2 (high) strikes a strong balance between performance and token efficiency
- MIT-licensed open weights
- Same API pricing as GLM-5.1

Tech Blog: https://t.co/LAsxUdN0JZ
Weights: https://t.co/g0A1C4UWx4
API: https://t.co/Kc3E22cbN7
Coding Plan: https://t.co/Nk8Y98HNhU
Chat: https://t.co/WCqWT0qCQb

[Przemek Chojecki | PC @prz_chojecki]

Kimi 2.7 ranked 2nd after Fable 5 and before GPT-5 xhigh

We have re-run our ErdosBench smoke test on 14 problems with Kimi 2.7, Qwen 3.7 Max, Grok 4.3 and compared it with the top performers from previous runs.

Kimi 2.7 is amazingly good. More below.

[Sakana AI @SakanaAILabs]

AIを作るAIを作る：RSI Lab始動

https://t.co/Tvc1zQyKkk

Sakana AIは、再帰的自己改善(Recursive Self-Improvement、RSI)に取り組む専任の研究グループ「RSI Lab」を、東京で立ち上げます。RSIは、AIがAIそのものを作る仕組みです。

この2年間、私たちはLLM-Squared、Darwin Gödel Machine、Shinka Evolve、ALE-Agent、Digital Red Queen、そしてThe AI Scientistといった研究を積み重ねてきました。いずれも、エージェント用途のために設計されたモデルが研究を自動で行うAIを生み、そのAIがさらに優れたモデルを生み出していく、というひとつの循環に向けた歩みです。

自己改善型AIという発想は、いまや私たちだけのものではありません。2026年に入ってRSIは大きな潮流となり、この考え方を掲げる組織が世界各地で相次いで立ち上がっています。そのなかでSakana AIは、創業以来、計算資源の量に頼らずにAIを開発する独自の方法を切り拓いてきました。

私たちがめざすのは、計算資源を際限なく注ぎ込むことなくRSIを実現することです。計算規模で世界最上位の国と張り合うのが難しい日本だからこそ、取り組む意味のある研究だと考えています。RSIの研究を責任あるかたちで進めるため、これまでの経験も活かしながら、コミュニティの皆様と知見を共有しつつ取り組んでいきます。

国内外から研究者・エンジニアを集め、Sakana AI RSI Labを組成していきます。この取り組みに、さまざまなかたちで関わってくださる皆様と協働できることを楽しみにしています。

[Google Magenta Project @GoogleMagenta]

MRT2 on @huggingface Spaces and pytorch/transformers!

Now you can play the demos just from your browser (with a bit of extra latency)

https://t.co/BwbbPWbrcq
https://t.co/KwthTXTJNq

[Sakana AI @SakanaAILabs]

Introducing DiffusionBlocks: Block-wise Neural Network Training via Diffusion Interpretation

https://t.co/c9AvsRKybj

What if we didn’t have to hold an entire neural network in memory to train it?

Standard neural net training optimizes all parameters jointly. As a result, the memory required during training grows linearly with the depth of the network.

In our #ICLR2026 paper, we propose DiffusionBlocks, a principled framework to train networks one block at a time, drastically reducing memory requirements while matching end-to-end performance.

With DiffusionBlocks, we split the network into blocks and train them one at a time, so you only need memory for a single block.

How? We explicitly assign each block a role: to move the representation a little closer to the target than the block before it did. That role turns out to be precisely what a diffusion model does, step by step. Each block only needs to optimize its own objective and can be trained independently.

We validated this across five different architectures:

• ViT
• DiT
• Masked diffusion
• Autoregressive transformers
• Recurrent-depth transformers

In each case, performance is competitive with end-to-end training while using a fraction of the memory.

This perspective also extends naturally to recurrent-depth (Looped) transformers, which apply the same network iteratively and normally require expensive backpropagation through time (BPTT). Viewed through DiffusionBlocks, we can replace those multiple iterations with a single forward pass during training.

Read our paper and code, to learn more.

Paper: https://t.co/CRj96VGYQn
GitHub: https://t.co/eNW0K9Xh8E
🐟

[Josef Chen @josefchen]

If the Founder of Hugging Face asks, you gotta do it. Models and dataset now live: https://t.co/9gY7AiQ9xs

Also built an explorer:
https://t.co/2wSMBS5tQ7

[Variety @Variety]

Peter Jackson on AI being used in film:

“I don’t dislike it at all. I mean, to me, it’s just a special effect.”

[Proton VPN @ProtonVPN]

@vxunderground they're saying I do what?

[Proton VPN @ProtonVPN]

Every time Proton VPN receives a data request, we *politely* decline, but if push comes to shove, we have been known to send out blank A4 pieces of paper titled "user logs"

https://t.co/QtkzZNs2Ft

[〽️att Greenberg @McGreenBeats]

Three.js can do everything now.

VFX and audio sync handled in browser. Premiere for the final assembly. #threejs #op1 @jugendingenieur