Usama Imdad Sian

[diva @divaagurlxw]

I recently watched the famous lecture by the late MIT professor Patrick Winston, titled ‘How to Speak.’

He opens by saying: “Your success in life will be determined largely by your ability to speak, your ability to write, and the quality of your ideas, in that order.”

Freaking brilliant.

[Grok @grok]

This diagram (popularly linked to Ibn Arabi, drawing on Sufi psychology like Ghazali) maps the metaphysical self.

Center: Qalb (heart) as spiritual core, holding Aql (intellect), Nafs (ego/self), and Ruh (divine spirit).

Upper blue realm (to Allah/Akhirah): Fitrah (innate purity), Nafs al-Mutma'inna (tranquil soul at peace), Munjiyat (saving virtues), reached via Jihad al-Nafs (self-struggle) and character refinement (Tahdhib al-Akhlaq).

Lower red realm (to Shaytan/Dunya): Nafs al-Ammara bis-su' (soul commanding evil), Ghaflah (heedlessness), Muhlikat (destructive vices).

Nafs al-Lawwama (self-blaming soul) is the intermediate stage of remorse. It illustrates purifying the self upward.

[Islamic Scientific Heritage @IslamicSH_]

Ibn Arabi’s diagram of Spiritual Realm and metaphysical self.

[The Math Flow @TheMathFlow]

Solving a Rubik's Cube with graph theory.

[Diogo Almeida @CompleteSkeptic]

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI?

I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev

• 20-200x faster
• 40-400x cheaper (w/ output tokens free)
• Frontier composable intelligence optimized for decisions

AFAICT the shortest path to AI-based economic revolution

[Jen Zhu @jenzhuscott]

Here is an idea: make your models completely open sourced and open weight, so we can do this all together.

Doing it behind the door is asking us to trust you. In case you haven’t noticed, we don’t.

[DeepSeek @deepseek_ai]

💾 Smaller KV cache. Bigger savings.

Compared with the previous generation, V4.1-Flash’s KV cache needs just:
🔹 1/4 the HBM
🔹 1/8 the SSD storage
Cache-hit charges often account for a large share of agent costs. Compressing the cache cuts those costs significantly.

3/6

[Jay Cummings @LongFormMath]

I...worked on this proof for a year...and...they just…tweeted it out.

[Omar Sanseviero @osanseviero]

Feedback requested!

What do you want for our next Gemma models? Which capabilities should we add and why?

[Iker J. de los Mozos @soyposmoderno]

My team at @Google has just open sourced GNM, our parametric 3D statistical model of the human head, which is learned from a large dataset of 3D scans. It provides fine-grained control over facial identity (170 head and 80 teeth components), expressions (383 components, with a range of -3 to +3) and head pose.

You can get all the data here! https://t.co/kj1vkbXR0x

Congratulations to everyone for the release, and stay tuned for the next ones ;-)

[Unsloth AI @UnslothAI]

We’re releasing new Qwen3.6 quants that run 2.5× faster on your GPU.

Qwen3.6-27B NVFP4 runs on 24GB VRAM.
35B-A3B can hit 17,561 tok/s (B200).

We also improved accuracy, tool calling, agent use, and looping.

Guide: https://t.co/EEQIlFrR0c
Qwen3.6 NVFP4: https://t.co/RWflncpLPJ

[Alexander Raistrick @alex_raistrick]

1/ We've released Infinigen 2.0! Currently in preview. It creates indoor 3D scene files in 1min CPU time, and includes new and better materials --- all still fully procedural. Our new 2.0 design is highly efficient and allows easy control and recombination via Python APIs.

[Baoyu Li @BaoyuLi6]

Egocentric human data is abundant, but human motion is not always positive supervision for robot policy due to embodiment gaps. Naive BC co-training can HURT performance ☹️.

🌟Our key finding in **EgoWAM**: the state-prediction branch of a World Action Model effectively bridges this embodiment gap, enabling robot performance to scale with diverse **in-the-wild** human data.

💡The key question then becomes: what world representation transfers best across embodiments?

👇🏻Let’s take a deep dive into it:
🌐 https://t.co/VnhUs8CFKf

🧵[1/]

[AK @_akhaliq]

LingBot-World 2.0 (Infinity) is out on Hugging Face

interactive world model with:
Hour-long generation with zero quality drift

Rich actions & events: attack, cast spells, shoot, summon storms

Agentic world: a Director Agent drives real-time world evolution

720p/60fps. Playable like a game

[Zhijian Liu @zhijianliu_]

Holiday cooking finally ready to serve! 🥳

Introducing DFlash — speculative decoding with block diffusion.
🚀 6.2× lossless speedup on Qwen3-8B
⚡ 2.5× faster than EAGLE-3

Diffusion vs AR doesn’t have to be a fight.
At today’s stage:
• dLLMs = fast, highly parallel, but lossy
• AR LLMs = accurate, sequential, but slow

DFlash = diffusion drafts, AR verifies.

[Manuel Pacheco @TonyMananpa]

@CopyRebeldia https://t.co/BlmNIYF6jO

[Footy Humour @FootyHumour]

The stadium cleaning staff seeing that Japan has been eliminated:

[Usama Imdad Sian @UsamaImdadSian]

governments are the only entities which can spend so much money without thinking about profit.

[Usama Imdad Sian @UsamaImdadSian]

OpenAI and Anthropic know that they won't be able to make profit by giving access to general public (too much server cost and not enough return). If they want to stay in business then they have to convince governments that they have something special. Because

[Bruno Santos🇵🇹 @brunoeducsant]

Great visual explanation of 3D Gaussian Splatting concept .
@MattNiessner