Deepak Subramani

PhD, @MIT. BTech, @iitmadras. AI, ML, Climate, Upskilling, Education. Views are my own.

[Deepak Subramani @deepakns]

@cvkrishnan lol. Same question I had

[Weirui Ye @Weirui_Ye]

I scraped all 38,101 ICLR 2027 submissions and counted the names before the colon.

LLM has a naming crisis.

🏆 The leaderboard:
TRACE: 47 papers
PRISM: 42
PACE: 22
SCOPE: 21
SAGE: 17
MOSAIC: 16
COMPASS: 14

750 names are used by 2+ papers. That's 2,209 papers sharing a name.

Best parts:

• Two papers called TRACE both fix VLM hallucination. Same name, same problem, different authors.

• Four papers called PRISM do time series. Good luck, reviewers.

• "PACE: Discovering Options via Self-Distillation." None of P, A, C, E appear in the title. They picked the name first.

• 47 TRACEs cover quantum computing, materials discovery, seizure localization, carbon emissions, traffic control, and circuit design. TRACE is everywhere.

• The formula: 5 letters, looks like a real word, sounds cool. CAST. PACT. CRAFT. GRAFT. FORGE.

Tip for ICLR 2028: search OpenReview before you name your method 😂

Follow for more data-backed ML nonsense.

[Deepak Subramani @deepakns]

lol

[Deepak Subramani @deepakns]

આવજો અમદાવાદ! સ્વાતિ સ્નેક્સનો સ્વાદ, અગાશીયેની શાહી થાળી, હયાતની મહેમાનગતિ, દાસના પોચા ખમણ અને દક્ષિણ ભોજનમનું અસલ ભોજન. સાયન્સ સિટી, કાંકરિયાની સાંજ અને ગાંધીનગરની હરિયાળી. મહેનતુ યુવાનો અને શેરીઓમાં હાસ્ય ને સ્મિત. પેટ અને દિલ બંને ભરાઈ ગયા. ફરી મળીશું!

[Deepak Subramani @deepakns]

Kuriachan naatil evideya

[Trung Phan @TrungTPhan]

If Jensen steps down, Kurian and Garman are def top two Nvidia CEO candidates (especially because Sundar and Jassy aren’t leaving their jobs).

Both manage $100B+ capex spend. Know the entire AI ecosystem. And honeypot risk is low based on fact they both pay for Linkedin Premium.

[Deepak Subramani @deepakns]

Before going to IIT, I thought I was 1. But then I saw real 1 at IIT, then to cope, I went the route of 2. That took me ahead and now I am a full fan of 2.

[Akash Sampurnanand Pandey @late_bloomer_7]

Two kinds of students crack IITs

1. So effortlessly brilliant, they put Ramanujan to shame.

2. So hard-working, they put donkeys to shame.

I used to be a fanboy of type 1 when I was preparing.

Now, after 10 years of graduating from IIT, I see the second type a lot more successful

[Deepak Subramani @deepakns]

I think only thing that will make sense is allowing people to submit only if they have previously reviewed as per community standards. Being reviewer in a venue where you are an author with limited slots is CoI

[Jessica Hullman @JessicaHullman]

It really is pointless and unnecessary to allow external reviewers to use AI as they want. There are obvious better alternatives.

By now, everyone should already be using AI to check their work before submitting. So what's the point of reviewers meat proxying to send more AI-generated feedback? Seemingly none.

What happens if we ban AI use in peer review? The main risk is to the imagined role of peer review as a basic soundness filter. But this is much better addressed by the venue having a centralized approach that uses validated AI tools to check that a paper is what it says it is than by individual reviewers feeding papers to LLMs using basic prompts. Such checks might include things like whether the results can be reproduced from the data/code, whether claims can be traced back to some provided evidence, whether there are TODOs left by an LLM to replace table results with non-fake data, etc.

Some of these can be used to determine if a paper should be desk rejected. Anything that is more controversial can be surfaced for reviewers.

This will shift what reviewers pay attention to, but this can be addressed by: 1) making clear to reviewers that their role is to identify what these checks miss and to weigh them, 2) being intelligent about how you select the set of LLM checks that are surfaced. See e.g., our work on learning LLM-implemented checklist that avoid adding more bias than they reduce variance. https://t.co/Np9qW1upuz

What happened with ICLR being permissive with reviewers using AI freely is seeming pretty inexcusable.

[Deepak Subramani @deepakns]

Non serious lottery seeks are plenty

[King Maximus @Pe4seve9ra5ceup]

Ashwini ji is serious💀

[Deepak Subramani @deepakns]

Such a development is needed in the South. Smart city mission is good, need larger scale

[Deepak Subramani @deepakns]

Now I understand why the PM has so much support in Gurjar Pradesh. Karnavati is a top class world city. Wide roads, modern amenities, so many open spaces and things to do. Just if the weather was good…. Had a fun time at Kankaria Lake, Science City, Riverfront etc

[Deepak Subramani @deepakns]

@SAnsumali @abhas_rewcie Maybe 90s site is needed to prevent hacking? (As modern hackers don’t know old programming ;P )

[Ramkumar Comic @ramkumarcomic]

Annachi takes on AI coding 🥲 No AI coding took on Annachi 😂 Catch Retired NOT Hurt in Chennai on Nov 21 and Bengaluru on Dec 5 - the last two shows of 2026.
Vaanga Makkale! ❤️ 🇺🇸 🇨🇦 🇦🇺 Coming soon 🙌🤗

[Deepak Subramani @deepakns]

@krshubham1708 Using Argon? It’s your word vs other… don’t believe it

[Sashikumaar Ganesan @SashikumaarG]

Something we’ve built from the ground up. Tomorrow, we share it.

Supported under the IndiaAI Mission (MeitY) Foundation Model initiative.

#ZenteiQ #IndiaAI

[Deepak Subramani @deepakns]

For almost all X and LI interactions, I feel like saying Suraj’s dialogue about Murthy Annan: “He didn’t use the elephant, nor did he strike the king. Yet, how confidently he made the move and say impresses me” from Jailer 2.

[Deepak Subramani @deepakns]

@otherme101 Ok. Let us do a thought exercise. We get all the revenue, then we develop our city to world class level, all "dehaati" state peoples move here. Our Bengaluru core has about 20k people per sq km. Looking at the development, if more people move, it will be 40k. Do we want that?

[Deepak Subramani @deepakns]

Actually there is another angle to this story. If rich states keep all their winnings and grow, then it is a problem for them unless border control is put in place. Redistribution and quick growth of poor states is essential.

[Kannadahelp ಕನ್ನಡ ಸಹಾಯ @kannadahelp6]

Kannadigas check this!

There is no answer with them for Nikhil Kamat question! This is the question we r asking daily, what’s the incentive with Indian union in life? If we do better in life you snatch and feed subhumans in north India. We are leading life with ZERO benefits! 😡

[Deepak Subramani @deepakns]

@pic_a_daily I don’t know about them, but the constituencies in those states must grow, else migration and over crowding in our Bengaluru

[Deepak Subramani @deepakns]

You know that VB is fast because they get priority over other trains. Of all trains are VB, nobody gets priority on our congested network @cvkrishnan sir

[Krishnan @cvkrishnan]

Overnight T-18s if done right on the 800-1600 km segment would be a runaway success. There’d be demand for 2-3 such trains(spaced an hour apart) between cities within few years.

[Bojan Tunguz @tunguz]

Yup, it's the new form of performative busywork.