Ayush Thada

A constant in a pool of variables. Proud Nerd and Geek. Analogue by birth, Digital by design. Love Math, Computers and Earth. #(B|A)yesain

[Ayush Thada @itsayushthada]

"Geometry is not the stage on which events unfold, it is the shape of the action itself."

This is probably one of the best mathematical quotes that I've ever read. Check the associated video by @JasonA_Ross which is really amazing btw.

https://t.co/JRFJkh7tEY

#geometry #math

[AilaunchX @Ai_Tech_tool]

🚨 Anthropic just showed a 27-minute workshop on how to actually do prompts for Claude.

Taught by the people who built it.

Free. No registration. No paywall.

I've seen $300 courses that don't cover what they teach in the first 8 minutes.

Watch it and bookmark it now.

[Pavlo Molchanov @PavloMolchanov]

What if you could take three completely different model families… and distill them into one tiny model? 🤯

📜 Paper: https://t.co/K2iKD4xFvp

MOPD (Multi-Teacher On-Policy Distillation) has become a standard procedure in post-training. We already distill multiple specialized variants of the same model into a single set of weights.
But what if we could go further - and distill models from entirely different families? Turns out, it is possible.

Today we’re releasing a paper on cross-tokenizer distillation - our first steps in this exciting direction. 📄

We distilled Qwen3-4B, Phi-4-Mini, and Llama-3B into Llama-3.2-1B.
MMLU jumped from 32.05 → 46.32 when using multiple teachers. 📈

The team is now working on Nemo-RL integration so the community can try this method in their own settings. Plus, we are scaling experiments up. 🚀

[Emerson S @Em_Nomadic]

Wanna learn robotics?

@tnkrdotai is GitHub for robotics. One place. Find real projects. Learn from real builders.

Come build with us 🌎

[Rashree @FIR31415]

every engineering college in India should quadruple the funds disbursement for the robotics, hardware labs/clubs!
we will see magic!!!

[Sarvam @SarvamAI]

We are excited to announce that Sarvam is partnering with @PixxelSpace to power the AI backbone of India's first orbital data centre satellite.

This is a first for the country, with India-built AI models running on an India-built satellite and both training and inference happening directly in orbit, without any dependence on foreign cloud or ground infrastructure.

[Mr Melancholy @chakravartiiin]

> 100 year old Toilet Manufacturer company

> Makes one of the most critical parts needed to manufacture AI Chips

Japan is in another league

[Jamie Simon @learning_mech]

did you know that with a few modifications, you can get the Ising model to simulate cells fighting to the death? one of my favorite side projects of all time:
https://t.co/EkuplsbV8o

[Sanjay Srivastava @Sanjay_Sriv]

Japan & India are turning quantum tech into geopolitics + business. This isn’t just a research pact, it’s the start of an Indo-Japanese deep-tech corridor. 🇯🇵🤝🇮🇳

Signing today. The agreement is expected to accelerate quantum computing collaboration between researchers and businesses in both countries.

This could help India gain early access to quantum hardware, expertise, and industrial use cases before the technology matures globally.

Japan brings hardware and know-how, India brings scale and ambition. Quantum is the bridge.
@PMOIndia @JPN_PMO

https://t.co/qbUKpErXcm

[Chao Ma @ickma2311]

Efficient AI Lecture 10: MCUNet and TinyML
(CNN On Microcontrollers)
An inspiring lecture on how to run neural network inference on extremely resource-limited hardware.

On microcontrollers, the bottleneck is often not the computation. It is whether the model can fit and run at all.
🔹 Flash stores the model weights
🔹 SRAM stores intermediate activations
🔹 Early CNN layers can create the peak memory bottleneck
🔹 Patch-based inference reduces activation memory

The lecture is about designing the model, memory usage, and inference schedule together so AI can run on tiny devices.

My note:
https://t.co/XotSpWzBwe

[Sebastian Raschka @rasbt]

Here is a 2nd batch of April architecture drops. What a month!
- Ant Ling 2.6 1T
- Minimax M2.7
- Xiaomi MiMo V2.5
- Poolside Laguna XS.2
- Tencent Hy3-preview
- IBM Granite 4.1

[Oxford Mathematics @OxUniMaths]

Pure mathematicians tell us that our 30 billion or so credit, debit and prepaid cards are only safe because of number theory, and prime numbers in particular. But how?

James Maynard takes your details in his Sophie Germain lecture. Watch it all: https://t.co/uIfMAn9a8I

[Roland Bouman @rolandbouman]

Compiler construction is one of the oldest, best understood CS fields. It's decades of work by the brightest minds, and it's grounded in logic, informed by experience and strictly deterministic.

Comparing that with LLM-based coding agents is just wrong.
https://t.co/4EVmjMIAwn

[Yann LeCun @ylecun]

https://t.co/cfwFhKL5Aw

[Vaibhav Sisinty @VaibhavSisinty]

Wait. The world's biggest toilet maker is now an AI company? 🤯

I had to read this twice.

TOTO, the Japanese company famous for heated, music-playing, self-cleaning toilets, just became one of the quietest winners of the entire AI boom.

Their stock jumped 18% in a single day. The biggest one-day gain in their history.

Here's what actually happened.

For 40 years, TOTO has been making pizza-sized ceramic plates called electrostatic chucks.

These plates sit inside the giant machines that manufacture memory chips.

They hold each silicon wafer perfectly still while plasma carves microscopic patterns into it.

The plates have to be flat to an atomic level, survive plasma without
warping, and last thousands of uses.

Only a handful of companies in the world can make them. TOTO is the second largest.

How did a toilet company become one of them?

Because making a premium toilet requires the exact same skill.

Perfectly smooth ceramic, fired at 1,200 degrees, that lasts 30 years without cracking.

Same skill. Different end product.

Now here's where AI comes in.

Every AI model on earth runs inside data centers.

Those data centers need two things in massive quantities. GPUs to do the computing, and memory chips to store the data the GPUs are working on.

The memory chips are mostly NAND flash. The same kind in your phone, but at a scale most people can't imagine.

NAND chips are made by etching billions of microscopic patterns into silicon wafers.

The wafers are held in place by TOTO's ceramic plates while that etching happens. No plates, no chips. No chips, no AI.

Every time OpenAI trains a new model.

Every time you ask Claude a question. Every time an AI data center comes online.

There's a TOTO ceramic plate somewhere in the supply chain that made it possible.

The numbers are wild.

→ Their ceramics division grew 34% last year

→ Margins jumped from 9% to over 40%

→ Profit from chip components just passed profit from toilets for the first time ever

A toilet company is now a critical link in the AI supply chain. And the market just realized it.

Here's the takeaway nobody is saying.

The biggest winners of the AI boom aren't the names on the headlines.

They're the companies sitting quietly in the AI supply chain, doing one thing better than anyone else for decades.

You don't need to build AI to win in the AI era.

You just need to find your spot in the AI supply chain and own it better than anyone else.

[ISRO @isro]

Congratulations to GalaxEye on the successful launch of Mission Drishti!
A significant milestone in India’s space journey, with
world’s first OptoSAR satellite and India's largest privately built satellite, advancing all-weather Earth observation capabilities.

ISRO is proud to support this achievement by enabling access to its satellite testing facilities fostering innovation and handholding India’s emerging private space sector.

Best wishes to the team at @GalaxEye for continued success in strengthening the nation’s space ecosystem and advancing cutting-edge technology.

[CG @cgtwts]

Two Anthropic engineers just spent 24 minutes revealing claude code features most people don’t even know exist

Most people will scroll past this, which is exactly why you shouldn’t.

[Hamzé 🦀 @Hamzeml]

Python made AI accessible.

Rust can make parts of AI understandable.

That’s the bet behind Category Theory for Tiny ML in Rust.

We’re building tiny ML systems from first principles using:

Rust types
typed transformations
composition
training loops
category theory as an engineering tool

Not abstraction cosplay.

Executable structure.

Working draft. Public feedback welcome.

[Sam Khawasé @samkhawase]

What if India built GitHub?

[priyanshu.sol @priyanshudotsol]

someone wrote a 680 page interactive book on cs algorithms