Arun Prakash

AI Research.
AI4Bharat, IITM, India.
Advaitin

[Arun Prakash @a_arun283]

@aravind Using self-awareness to prove consciousness is flawed. Alexa and Siri are "self-aware" (respond to their names)—but not conscious. A newborn is conscious before Ahamkara (the self) develops in buddhi. AI operates at the level of buddhi (collective human buddhi). Not conscious!

[Arun Prakash @a_arun283]

@aravind Using self-awareness to prove consciousness is flawed. Alexa and Siri are "self-aware" (respond to their names)—but not conscious. A newborn is conscious before Ahamkara (the self) develops in buddhi. AI operates at the level of buddhi (collective human buddhi). Not conscious!

[Arun Prakash @a_arun283]

@aravind AI can be intellectual, able to interact with its surroundings with its given identity. But it cannot become emotional; it will never get bored; it will never feel inferior/superior; It will never bother to ask, " Who am I? " or " What is existence"?

[Sridhar Vembu @svembu]

The economic and demographic effects of corruption.

Cost of land in our urban areas is far higher than what our GDP per capita would dictate. The ratio of land value to per capita GDP is probably higher in India than anywhere else. As an example, land prices in Chennai or Bengaluru rival that of cities like New York which has a vastly higher per capita GDP.

The key reason?

First, vast sums of political corruption money is parked in real estate. This raises real estate prices and high real estate prices affect everything downstream.

Second, corruption in building approvals and the like - the famous DTCP - raises construction costs, on top of already higher real estate costs.

Third, corruption in private school regulatory compliance enforcement raises school fees.

Fourth, corruption in private hospital regulatory compliance enforcement raises health care costs.

Fifth, household goods need sales outlets and those pay higher rents due to high real estate prices and construction costs.

So housing, education, healthcare and household goods - all of these now cost higher.

As a direct consequence, the economic burden on the average person gets worse. Young people, facing all these costs, postpone marriage, and postpone children or have fewer children.

That directly affects our demographics.

While this issue exists in many parts of India, Tamil Nadu, being the most urbanized of the bigger states, is particularly hit hard.

So corruption is becoming an existential threat to our society.

If you worry about the super-low birth rate in Tamil Nadu, way below replacement, understand that corruption raising our cost of living is one of the major causes, not the only cause, but a big one in our context.

[S Gurumurthy @sgurumurthy]

DMK destroyed the character of TN children now. We all grew up reading and memorising Avvayar's Athichudi and Konraivendan. Here you hear Konraivendan in this video. Will TVK which promises change reinstate this. Beware EVR will be very upset
https://t.co/qfDkoeKCI0

[Sridhar Vembu @svembu]

The DMK ecosystem is attacking me that the school issue I reported was a lie I concocted to help the TVK government. Let me state the facts.

We run two rural schools, both free NIOS schools under the Kalaivani Kalvi Maiyam umbrella, one in rural Tenkasi and the other in rural Theni.

The Theni school was originally started and run by a retired IPS officer, a honest and upright man. He built very nice facilities but had to shut down his CBSE school because the state government demanded too much money to issue the NOC. He told me that as a honest retired officer he did not have the money to pay and they would not issue the NOC otherwise.

He urged us to take over his trust so we could run our free NIOS school in the premises. That is how we started our NIOS school in Theni.

Unlike in Tenkasi, where I live, the Theni school faced occasional harassment from the DMK government because we did not have a state government registration, so we tried getting state government approval again (this time for the state board) and that of course would also cost money. So it was in a limbo.

Meanwhile in Tenkasi we wanted to build new school facilities (we were operating in make-shift facilities, the ones that came in frequent photos) and we applied for DTCP approval to construct the new buildings. Everyone who knows DTCP in our state knows what kind of corruption happened there under the DMK.

We waited patiently for DTCP approval for the new school buildings but the approval never came as long as the DMK was in power. The approval came automatically once the government changed.

It is this DTCP approval that I posted about in X. I want to once again appreciate the refreshing change.

Not only did the approvals come, government people told us not to pay money to anyone for any approvals. I have to appreciate this in public, having endured what we had endured before.

This is the "lie" that DMK wants to attack me on.

I do not need their certificate on my character.
They can examine their conscience and ask why a technology nerd like me who is mostly immersed in code would post these.

If they think I would back down by their vile personal attacks because I am a Brahmin or TVK Stooge or Sanghi or whatever, I will tell them this: unlike you dynasts I grew up with nothing. I studied in Tamil medium schools. I know how to live on nothing. I have dedicated the remainder of my life to make Bharat self reliant in technology while reviving our rural areas, the soul of our eternal sanatana civilization.

I will not be intimidated by their attacks. I am unafraid of death, why would I be afraid of the mere DMK?

And if they had any conscience they can return the money they looted (they know there they keep it) and then they can attack my character.

I will now go back to optimizing the code of our compiler!

[Sridhar Vembu @svembu]

This is big: all access to Mythos and Fable AI models disabled for everyone outside America.

First thoughts:

1. Technology is the ultimate weapon. National sovereignty, national security, all of it is now about technology.

2. Globalization is dead and Bharat must find her own way ahead.

We must keep these two ideas in mind.

What can our government do right now? Ensure that orgs in India embrace smaller models, both Indian and Chinese open source ones. With a bit of effort, we can make them work. Anyway, why pay money to people who don't even want to sell to you?

We must deepen our R&D. Sarvam has been on it and we have been on it but remember that the latest models cost not only huge GPU budgets to train, the GPUs themselves are restricted. So we can't afford the scale of money (of the order of $100+ billion to even get in the game!) and even if we could come up with the money, we can't get all the GPUs. I would not like to ask the government to fund tens of billions of dollars on this anyway - the money has far better uses.

Zoho has been pursuing alternative R&D approaches that are far, far less expensive but by its nature cutting edge R&D takes time and we are patient. I am confident we will get there.

Any remaining people in India who have delusions about globalization should wake up now.

[Sridhar Vembu @svembu]

As of 2026, Tamil Nadu is the only state in India that does not have any Jawahar Navodaya Vidyalayas (JNVs). As the name indicates, this policy was formulated when Congress was in power and the BJP has continued it.

It is generally a bad idea for one state to keep resisting policies every other state, including non-BJP states, accept.

DMK pursued that arrogant policy and lost.

Given that the Congress is in the ruling alliance in Tamil Nadu, I hope they push for Jawahar Navodaya Vidyalaya schools in the interest of our Tamil Nadu students.

Accepting Jawahar Navodaya Schools would not lose a single vote and in fact may win many votes from grateful parents. Pragmatism in such matters is politically wise.

[Andrew Ng @AndrewYNg]

New course on serving LLMs efficiently -- how do you serve models to many concurrent users at low latency and reasonable cost? This short course is built with @RedHat and taught by @cedricclyburn.

Efficient LLM serving requires efficient memory management. A 70B-parameter model takes ~140 GB just to load the weights. On top of that, every active request needs its own chunk of GPU memory, the KV cache, to store the token context it has built up so far. In this course, you'll learn to reduce a model's memory footprint with quantization and serve it using vLLM, which handles many concurrent requests efficiently through smart memory management.

Skills you'll gain:
- Quantize a model and measure the accuracy tradeoff
- Serve a model with vLLM and watch it handle concurrent requests efficiently
- Benchmark your deployment and make informed tradeoffs between speed, cost, and accuracy

Join and learn to serve LLMs efficiently:
https://t.co/x04xMbFlkO

[Sridhar Vembu @svembu]

It is such a fraudulent thing to claim that Tiruvalluvar is not Hindu. This attempt to deny our own spiritual heritage must be fought.

[Elon Musk @elonmusk]

😂

[Arun Prakash @a_arun283]

There are many people from Bharat who contributed immensely to science. Every teacher must share lesser known stories like these to all students in their class.

[Swarajya @SwarajyaMag]

One of the most important tools in modern biology is used every day in labs worldwide.

Most people have never heard of the man who created it.

His name was G.N. Ramachandran. 🧵

[Arun Prakash @a_arun283]

@svembu AI can be exponentially constructive or destructive. If students know upfront that they cannot use AI during exams, they naturally pivot to using it for actual learning

[Sridhar Vembu @svembu]

One company I hugely admire is NPCI, and every time I use UPI, I think of their work and the massive impact! So it is an honour to do the podcast with their CEO Shri Dilip Asbe. Thank you Dilip-ji🙏

[Sridhar Vembu @svembu]

AI inference cost (which we pay in dollars) may rival our oil import bill and blow up our current account deficit. Great post on that below.

What is the solution? I believe that high developer productivity can be achieved without the high AI inference bill. We have to invent our way out of trouble. Stay tuned.

[K.Annamalai @annamalai_k]

Days after DMK MLA Thiru Udhayanidhi Stalin spoke about eradicating Sanathana Hindu Dharma on the floor of the legislative assembly, TVK MLA Thiru VMS Mustafa came in support of Thiru Udhayanidhi Stalin’s remarks. Not only is this very unfortunate but also it shows the level of hatred they harbour in their minds towards an eternal Dharma that helps us to lead a meaningful life!

DMK & TVK, in their very short history, have proven to the people that they are two sides of the same coin. DMK has forgotten what their constant attack on Sanathana Hindu Dharma has brought them to. TVK will also soon remember.

If either DMK or TVK had the courage of conviction, they should have openly declared during their election campaigns that targeting Sanathana Hindu Dharma would be central to their politics. Instead, they hid behind secularism & theatrics, only to reveal their true intentions after seeking votes from the very people whose faith they now insult with arrogance and impunity.

Please be reminded, Sanathana Hindu Dharma is not your punching bag! I sincerely hope their remarks will not go unpunished by our people.

[Arun Prakash @a_arun283]

சனாதனம் என்றால் எக்காலத்திலும் இருப்பது என்று பொருள். சனாதன தர்மம் என்றால் எக்காலத்திலும் இருக்கும் தர்மம் என்று பொருள். உதாரணமாக "அஹிம்சை". யாரும் எக்காலத்திலும் தன்னை ஒருவர் இம்சை படுத்துவதை விரும்புவதில்லை என்பதிலிருந்தே அது "சனாதனம்" என்று அறியப்படுகிறது.

[Sridhar Vembu @svembu]

When Udayanidhi Stalin repeatedly attacks "sanathanam" in Tamil, why is there not more outrage among the highly religious Tamil Hindu population?

I believe it comes down to language.

In Tamil language, the word "sanathanam" (eternal in English) is not in regular use. Most Tamil people do not know this word and I did not know it myself as a child.

On the other hand the word "dharma" (spelled as "dharmam" or "dharumam") is common in Tamil and there is a district named Dharmapuri in Tamil Nadu. There is also the word "aram" used in Tamil with the same meaning as dharma.

Interestingly, in Hindi, the word "dharam" means "religion" in general and Hinduism is "hindu dharm" or "sanatan dharam" and Christianity is "isai dharam" in Hindi.

But in Tamil usage, the word "dharmam" would effectively mean "Hindu dharam" in Hindi.

Now if Udayanidhi had attacked "dharmam" in Tamil (which effectively means Hinduism) he would be widely criticised for proclaiming adharmam.

That is why he is picking the word "sanathanam" to attack because that word is not widely known in Tamil.

With all his "sincere" effort, he is making the word known in Tamil too and he has made the already huge anti DMK vote in Tamil Nadu even bigger. The DMK has never won a majority on its own ever, because of that huge strong anti-DMK current. They established that current with their constant attacks on Hindu deities. The AIADMK never did that and they harvested the anti-DMK current well.

Now, in spite of the TVK being the new anti-DMK party and in spite of the AIADMK splitting the anti-DMK vote (see how I worked it!) and in spite of the DMK spending extraordinary amounts of cash, in spite of the DMK having a strong alliance and the TVK having no alliance, the DMK lost the election.

Udayanidhi wants to ensure the DMK would never come back. May the eternal sanatana dharma grant his wish 🙏😉

[Swarajya @SwarajyaMag]

A $35,000 Ukrainian drone destroyed a Russian Su-57 worth $54 million while it sat on its airbase. For every dollar Iran spends on a Shahed, the US spends $28 to shoot it down.

This is the new economics of war, and India will need a serious answer to where it leaves us. 🧵

[Arun Prakash @a_arun283]

The Universal Policy: What Reinforcement Learning Teaches Us About Life

Reinforcement Learning (RL) is the science of discovering a policy (a mapping from situations to actions) that maximizes an agent's long-term cumulative reward in a given environment. Today, it powers the post-training stage of Large Language Models, aligning their responses with human values and preferences.

We can apply the same lens to human life. Each of us is, in a sense, an agent. Every day we take actions, hoping to maximize some long-run reward. The nature of that reward differs from person to person, but the underlying objective tends to converge on the same thing: to live in happiness and peace — not merely at the end of life, but at every step along the way.

This raises a fascinating question. Does an optimal policy exist for humans navigating the complex environment we call life? If it did, we could make every decision without anxiety about the trajectory ahead. By definition, optimality would guarantee a high-reward outcome with zero regret.

But here we run into a wall. Each person's life-environment is drastically different. Borrowing the policy of someone successful rarely transfers cleanly, as no two lives are identical, and the environment itself is constantly evolving. On the surface, it seems impossible that a single policy could fit everyone.

But a universal policy does exist. It is not complex! It is so elegant that it collapses into a single question at every decision point:

Is it dharmic?

If yes, act. If no, don't.

That's the entire policy. Just as RLHF aligns a language model's values with those of humans, dharma aligns an individual's values with those of the universe. Universal in scope, optimal by design, and yet — paradoxically — one of the hardest policies for humans to actually follow.