Tejasri

Star gazer & beach lover

[Tejasri @tejasri19]

Organ on chip and agentic AI powered autonomous personalized drug discovery is revolutionary!

[Kolli Sarath @kollisarath]

https://t.co/Z9lKhnOGk6

[Tejasri @tejasri19]

@kollisarath Organ on chip and agentic AI powered autonomous personalized drug discovery is revolutionary!

[Karishma Kaushik, MBBS, MD, PhD, Mom @_KarishmaK]

Most women PhD holders in STEM in India find themselves unable to access long-term, lucrative, and prestigious research jobs.

On #InternationalWomensDay, I write for @the_hindu on WHY inspite of boasting of the largest number of women STEM graduates - social, structural and systemic factors preclude the full and sustained participation of women in STEM research in India. 📝👩‍🔬

https://t.co/r3rOoeGS1D (behind a paywall)

[Sebastian Raschka @rasbt]

While waiting for DeepSeek V4 we got two very strong open-weight LLMs from India yesterday.

There are two size flavors, Sarvam 30B and Sarvam 105B model (both reasoning models).

Interestingly, the smaller 30B model uses “classic” Grouped Query Attention (GQA), whereas the larger 105B variant switched to DeepSeek-style Multi-Head Latent Attention (MLA).

As I wrote about in my analyses before, both are popular attention variants to reduce KV cache size (the longer the context, the more you save compared to regular attention).

MLA is more complicated to implement, but it can give you better modeling performance if we go by the ablation studies in the 2024 DeepSeek V2 paper (as far as I know, this is still the most recent apples-to-apples comparison).

Speaking of modeling performance, the 105B model is on par with LLMs of similar size: gpt-oss 120B and Qwen3-Next (80B). Sarvam is better on some tasks and worse on others, but roughly the same on average.

It’s not the strongest coder in SWE-Bench Verified terms, but it is surprisingly good at agentic reasoning and task completion (Tau2). It’s even better than Deepseek R1 0528.

Considering the smaller Sarvam 30B, the perhaps most comparable model to the 30B model is Nemotron 3 Nano 30B, which is slightly ahead in coding per SWE-Bench Verified and agentic reasoning (Tau2) but slightly worse in some other aspects (Live Code Bench v6, BrowseComp).

Unfortunately, Qwen3-30B-A3B is missing in the benchmarks, which is, as far as I know, is the most popular model of that size class. Interestingly, though, the Sarvam team compared their 30B model to Qwen3-30B-A3B on a computational performance analysis, where they found that Sarvam gets 20-40% more tokens/sec throughput compared to Qwen3 due to code and kernel optimizations.

Anyways, one thing that is not captured by the benchmarks above is Sarvam’s good performance on Indian languages. According to a judge model, the Sarvam team found that their model is preferred 90% of the time compared to others when it comes to Indian texts. (Since they built and trained the tokenizer from scratch as well, Sarvam also comes with a 4 times higher token efficiency on Indian languages.

[Sandeep Manudhane @sandeep_PT]

An article that'll help frame the issue well. Do read.

[IISc Bangalore @iiscbangalore]

We are excited to announce Open Day 2026! 📷

Visit our campus on March 7th between 9 am and 5 pm. Explore the exciting research demos, displays, exhibits and experiments!
Use the hashtag #IIScOpenDay2026 to share what you see!

Details: https://t.co/wSVFMp3Io4

[তন্ময় l T͞anmoy l @tanmoyofc]

THE DIRTY SECRET THEY WON’T TELL YOU

DATA CENTRES GUZZLING YOUR DRINKING WATER 💥

• India's Data Centres will suck 150 BILLION litres of water in 2025 → 358 BILLION litres by 2030 (BBC + Mordor Intelligence 2026). That's more than DOUBLE in 5 years.

• 60–80% of these Centres sit in HIGH WATER STRESS zones (S&P Global). Chennai, Bengaluru, Hyderabad, Mumbai — Exactly Where People already Fight for Every Drop.

• India: 18% of World Population, just 4% of Global Freshwater.
Per Capita Availability Crashing to 1,367 m³ by 2031 (Moody's).
163 million Indians STILL Without Reliable Safe Drinking Water (NITI Aayog).

• Jal Jeevan Mission? Extended to 2028 because they couldn't finish by 2024.
Only 81.66% Rural Households have Tap Water as of Feb 2026.
Villages dig Deeper Borewells… While Data Centres get Priority Municipal/Groundwater.

• AdaniConneX, Nxtra, Sify Quietly Pulling millions of litres — Some Reports show Groundwater use with ZERO transparency on exact DC share.

• One 1 MW centre = 26 million litres/year = water for 528 Households.
India racing to 5+ GW pipeline. Do the math.

They scream "Viksit Bharat" and "AI Superpower" but won't Admit: While Farmers & Families Queue For Tankers, Hyperscale Servers Get Unlimited Water To Cool Silicon Valley's Profits.

#IndiaAIImpactSummit2026 #TrumpTariffs #DonaldTrump #VoiceOfTruth

[Bloomberg Opinion @opinion]

Nowhere were the promises and hypocrisies of AI clearer than in India this week.

@cathythorbecke explains from New Delhi 🎥

[Tejasri @tejasri19]

Here's the draft release of dataset license by @SwechaFSMI @viswamai and others. The license is aimed at preserving dataset integrity and ensuring safety use.

[VISWAM.AI | A Joint Initiative of Swecha and IIITH @viswamai]

Following the roundtable on Understanding Trust & Safety in AI:From Code to Creativity,here is the draft release of the dataset license for public consultations addressing gaps in open-source licensing around data use,attribution&openness in the AI era.
🔗 https://t.co/nLFCGGJ6q7

[VISWAM.AI | A Joint Initiative of Swecha and IIITH @viswamai]

Following the roundtable on Understanding Trust & Safety in AI:From Code to Creativity,here is the draft release of the dataset license for public consultations addressing gaps in open-source licensing around data use,attribution&openness in the AI era.
🔗 https://t.co/nLFCGGJ6q7

[Snehil Saluja @mrsnhl]

A 1 crore earning techie's life has no value in Bengaluru.

And if you're not earning that much? Your life has even less.

My sister and her friend were driving home in my car. They stopped at a red light - the logical thing anyone does. A drunk driver in a mini-truck didn't feel the need to stop. He slammed into them instead.

I know he was drunk. She knows it. The highway police knows it. The truck owner knows it.

No arrest was made.

The truck driver never showed up at the station. The owner never showed up. Nobody cared. My sister and her friend - both injured, both terrified - kept going back to the station, back to the accident site, explaining what happened over and over, just trying to get a report filed.

I was in the US. All I could do was talk to them on calls, helpless.

Here's what the police told them:
"If nobody died, an FIR doesn't make sense."
"Just claim first party insurance."
"Third party insurance doesn't pay much anyway."

And then, quietly, one officer pulled them aside and told the truth: "These truck mafia bribe us. Nothing will happen."

Nothing happened.

The truck was KA04 AE6550. The police themselves said if they'd been on a two-wheeler, both would be dead.

We had 100% insurance from Reliance. Claim rejected. Reason? "Misrepresentation of facts." These two, even while injured, kept showing up to represent the facts. Reliance still found a way to deny them.

The law says if someone hits you from behind, the person behind is at fault. It was a red light. How does a truck driver not see that?

Trust me, this isn't about money. I'll manage the repairs and the medical bills. I have savings. And I have a decent credit score; I'll take a personal loan if I have to. That's not the point.

The point is this: my sister is afraid now. Afraid that anything can happen to her at any moment and there's no one - no system, no law, no institution - that will protect her.

But how do I tell her the world is supposed to be fair? How do I tell her to trust the system? How do I explain that the drunk driver walks free, the truck owner was never questioned, and the police pocketed their bribe and closed the file?

I can't say to any official, "What if this was your daughter? Your sister?" Because their daughters travel in cars with security escorts. They will never know what it feels like to be ordinary and unprotected.

So I'm saying it to you, an ordinary reader.

You're on the road. You stop at a red light. A drunk driver in a truck rear-ends your car. Your loved one is inside, terrified.

And then you learn: there is no recourse. None.

The truck owner pays off the cops. The insurance company rejects your claim. The system shrugs.

This is Bengaluru in 2025. This is India in 2025. This is what your life is worth here.

One more thing. The friend in the car? He's one of the smartest people I know. close to top 100 rank in IIT-JEE. AI engineer and one of the biggest data companies. At 23, he is valuable to be paid more annually than the cost of five such trucks, that too, in India. He's patriotic. He pays his taxes. He stays in India even though he constantly gets offers to move to the US.

This is the confidence our system gives to someone who is clearly an asset to this country. All this unfairness - for a drunk truck driver.

@blrcitytraffic @BlrCityPolice - tell me. I've always avoided raising fingers publicly. But what else can I do?

[TheLiverDoc™ @theliverdoc]

Can people in power/ authority kindly stop with this absolute BS on "fasting killing cancer cells" and citing religious nonsense to appeal to peoples emotions? Fasting is really quite dangerous for cancer patients in real life. Let me explain and bury this myth once and for all, especially for science illiterates like this guy.

[1] The most common argument is that fasting "starves" cancer by cutting off its sugar (glucose) supply. While it is true that cancer cells consume vast amounts of glucose, they are biologically aggressive survivalists.
If you stop eating, your body eventually switches to burning fat and breaking down muscle for energy. Cancer cells are highly adaptable; when glucose is low, many types of cancer can mutate to feed on other fuel sources, such as lactate, amino acids (from your muscles), or fatty acids. You cannot simply "starve" a tumor without starving the patient first.

[2] Fasting is dangerous for cancer patients because of cancer cachexia - a wasting syndrome where the body loses muscle and fat rapidly.
Cachexia is responsible for up to 30% of cancer deaths. Cancer puts the body in a hyper-metabolic state (burning energy fast). If a patient fasts, they risk accelerating muscle loss and weakening their immune system. A weak body cannot tolerate life-saving treatments like chemotherapy or radiation, nor can it fight off infections.

[3] Most claims about fasting curing cancer come from studies on mice or cells in a petri dish.
In a dish: You can kill cancer cells with almost anything (lemon juice, bleach, starvation, even shooting a bullet at it at close point or using a grenade to destroy the entire lab) because they have no immune system or body to protect them.
In a human: The biology is infinitely more complex. Human metabolism, hormonal fluctuations, and tumor micro-environments mean that what shrinks a tumor in a mouse often fails completely in human trials.

[4] Proponents often cite "autophagy" (the body's cellular recycling process triggered by fasting) as the cure. They claim it cleans out cancerous cells. Science shows that autophagy is a double-edged sword.
-In all people, autophagy is a normal physiological process - whether fasting or not, which help in cleaning up damaged cells.
-But in patients with cancer, once a tumor exists, cancer cells can actually hijack autophagy to survive stress (like chemotherapy) and repair themselves. In this context, fasting could theoretically help the cancer survive the treatment intended to kill it.

[5] "Cancer" is not one disease; it is over 200 different diseases characterized by uncontrolled cell growth.
Some cancers are driven by hormones, some by genetic mutations, and some by viruses.
A fasting protocol that slows down one specific type of breast cancer might have zero effect on pancreatic cancer, or worse, accelerate a different type.

The most lethal aspect of this misinformation is the delay in treatment. Cancer is a time-sensitive disease.
While a patient spends months trying to fast the cancer away based on religious or alternative advice, the cancer often metastasizes (spreads) to other organs. Once cancer spreads, it often moves from being curable to being terminal. Relying solely on fasting wastes the critical window where medical intervention could have saved a life.

Suggesting a single "ancient cure" for 200 complex genetic diseases is scientifically illogical...

...and generally stupid, as this video proves.

[FieldPhenomicsLAB @FieldPhenomics]

So, after global wheat, the new journey started ^^
We have never forgotten our staple food, rice, which is also a staple food in many Asian countries. We are honored to present the Global Rice Multi-Class Segmentation Dataset (RiceSEG):
https://t.co/YWPxAaafC9

[Math Cafe @Riazi_Cafe_en]

Terence Tao on working hard:

“Relying on intelligence alone to pull things off at the last minute may work for a while, but generally speaking at the graduate level or higher it doesn’t.

One needs to do a serious amount of reading and writing, and not just thinking, in order to get anywhere serious in mathematics; contrary to public opinion, mathematical breakthroughs are not powered solely (or even primarily) by “Eureka” moments of genius, but are in fact largely a product of hard work, directed of course by experience and intuition.

The devil is often in the details; if you think you understand a piece of mathematics, you should be able to back that up by having read all the relevant literature and having written down at least a sketch of how that piece of mathematics goes, and then ultimately writing up a complete and detailed treatment of the topic.

It would be very pleasant if one could just dream up the grand ideas and let some "lesser mortals" fill in the details, but, trust me, it doesn't work like that at all in mathematics; past experience has shown that it is only worth paying one's time and attention to papers in which a substantial amount of detail and other supporting evidence (or at least a "proof-of-concept") has already been carefully gathered to support one's "grand idea". If the originator of the idea is unwilling to do this, chances are that no-one else will do so either.”

Source: https://t.co/blJGOvOzCH
Image: https://t.co/j3wLCH5IrL

[Tejasri @tejasri19]

@balaji25_t @HeadRockShow Keep up doing the good work @balaji25_t 👏👏👏. Good to know we are from the same campus.

[Telangana Weatherman @balaji25_t]

Gave a nice and detailed podcast with @HeadRockShow . Shared so many facts about me, my personal lifestyle, journey as Telangana Weatherman & many more. Definitely do have a look guys 😉😀.

Thank you Jignesh sir, it was pleasure interaction with you 😀 https://t.co/gz8gtwNB6a

[Climate Change AI @ClimateChangeAI]

Another highlight of our #NeurIPS2024 workshop on Sunday will be a panel discussion on small vs big models in ML for climate impact, featuring Chris Bretherton, Anamika Dubey & @BDilkina.

Stream the whole workshop live & for free via our website: https://t.co/6cDIBuE7p6

[Tejasri @tejasri19]

More such programs are to be implemented to popularize science and its wonders.

[IISF 2024 @IISF_2024]

Join us for IIFS 2024, where we kick off the excitement with our Curtain Raiser and Outreach events! Discover the details of our dynamic venues and be part of the journey from November 30 to December 3 @CSIR_IND @csir_niist @vibhaindia @DBTIndia @IndiaDST @CSIR_IND #iisf2024

[Sasha Luccioni, PhD 🦋🌎✨🤗 @SashaMTL]

"In this new generative-AI paradigm, uncertainty reigns over certainty, speculation dominates reality, science defers to faith [...] while the technology consumes unprecedented amounts of energy."
An amazing piece from @_KarenHao about AI and climate:
https://t.co/DGzgo6i1SJ

[Krish Ashok @krishashok]

Had an amazing time with kids of class 8 from Chennai High School, Kottur thanks to @TeachForIndia and @Full_Meals. We spent an hour deconstructing a Chicken Biryani down to first principles - all the way from how the chicken’s ancestors were dinosaurs to how onions defend themselves using acid attacks to the eye. I urge everyone to go do this at least once. The kids are incredibly smart, high energy and open to learning.