Sanchit Sinha

Applied Researcher @CapitalOne | PhD @CS_UVA
Past: ML Research @MorganStanley | 2x Applied Science @AmazonScience

[Sanchit Sinha @snsanchit]

Accepted to #EMNLP2026 Findings! :)

[Sanchit Sinha @snsanchit]

Releasing a new version of Chart-RVR-3B, trained on curated hard chart reasoning tasks. This series of VLMs has been post-trained using GRPO on custom chart-specific rewards.
Vibe check here: https://t.co/4YQ4F6LTWU

Older version: https://t.co/7x1DNfsSNu

[Sanchit Sinha @snsanchit]

Interesting work!

[Oliver Li @oliveraochongli]

1/ Finally get to share what I’ve been working on this summer at @datologyai. We built DataSmith 🍊, an automated data researcher.

The idea is simple: scale test-time compute across multiple directions, then let agents synthesize their findings.

Blog: https://t.co/DFWT45uKLG

[Raj Dabre @prajdabre]

Citation counts are gameable, highly overrated and a very bad way to measure someone's research capabilities and impact. Comparison is often the thief of joy and I spent ages being miserable comparing myself to others. Instead the best thing is to focus on why you do research and enjoy the act of research itself. There is great fulfillment and joy in this.

Case in point, I once did a project on my native dialect of Marathi with my mom because I wanted to write a paper with her once in my life and unexpectedly got a best paper award for it. That paper has just 3 citations but I value them as much as all other citations on all other papers I have.

Remember the quote from Bhagavad Gita: You have a right to perform your prescribed duty (research), but you are not entitled to the fruits of actions (citations).

[Gon Huertas @GonHuertas]

If you're confused about what kind of writing gets flagged as "AI generated" by Pangram, this example may help.

Below is some human written text (to which I have deliberately added some typos). If you put this through Pangram, it'll be labeled 100% human.

"By and large, monetary authorities rely on a ssingle policy instrument: the reference nominal interest rate. The use of one instrument makes it imposible to target both inflation and output simultaneously, and creates an inherent tradeoff. The central bank's position on this tradeoff --how forcefully it is willing to pursue an inflation mandate in spite of the implications for economic activity-- defines the character of its monetary policy approach. We refer to central banks that react strongly to inflation, and that place a relatively small weight on output considerations, as being ``hawkish''.

How hawkish, then, are Latin American central banks? This section answers the question by estimating the monetary reaction functions, or ``policy rules'', of five of the region's largest economies for the 2003-2024 period. We find that they are hawkish indeed, responding to a ris in inflation with an aggressive increase in real rates, and prioritizing price stability to such an extent that output plays almost no role in their decision making.

This finding is consistent with explanations provided by central bankers in the literature, where it has been argued that central banks adopted a hawkish stance on inflation as a way to build monetary credibility, against the backdrop of the region's historical struggles with very high inflation.

Details on the sources of the data used in the analysis are provided in the appendix. We rely on monthly and quarterly time series from each country's central bank and national statistics office."

(cc @paulnovosad, @itaisher)

[Sanchit Sinha @snsanchit]

@gabriberton Interns are eligible for FT conversion. Student researchers are not. Google makes this distinction in the title.

[Sanchit Sinha @snsanchit]

@langer_han @PradyuPrasad What's frustrating is that ACs didn't really look at the rebuttal before posting the meta-review and leaned on the 100% Claude review. No one really reads the issues raised as well.
Getting disillusioned with the review process now.

[Sanchit Sinha @snsanchit]

@langer_han @zwcolin I agree; any optimization or fine-tuning paper should have stats. tests. Although AI asks for stats. tests on single-benchmark LLM runs as well, LOL.

[Sanchit Sinha @snsanchit]

@ScienceYael Don't hate the player, hate the game xD

[Sanchit Sinha @snsanchit]

GPT-5.6 is better than Opus 5 for my use cases. Good job @sama

[Sanchit Sinha @snsanchit]

If this happens…

[David Ornstein @David_Ornstein]

🚨 EXCL: Arsenal exploring move to sign Vinicius Junior from Real Madrid. #AFC interest early stage + clubs not yet in talks but idea approved at all levels. Much rests on contract process as no breakthrough yet & #RMFC do not want free exit @TheAthleticFC https://t.co/qYdS5vlfE6

[Sanchit Sinha @snsanchit]

@grok Discus vs Javelin world record in meters

[YouGov @YouGov]

Which do you think it is possible to throw further: a baguette or a naan bread?

A baguette: 51%
A naan bread: 36%

[Sanchit Sinha @snsanchit]

Do you have opportunities in the US?😁

[Karun Singh @karun1710]

We're hiring a Research Engineer at @Arsenal ⚽🔴⚪ to work directly with our Men's First Team!

We're building state-of-the-art AI models for the football domain. This role will focus on building the application layer for our research to advance coaching and analysis workflows.

[Sanchit Sinha @snsanchit]

Good night to everyone :')

[EBL @EBL2017]

Every coach can’t get what they want all the time, but oftentimes the alternatives in the market for Arsenal were cut-price deals like Trossard (who ended up being long-term solutions) or simply signing nobody at all.

Arsenal didn’t sign a forward for 4 straight transfer windows after Gabriel Jesus joined (aside from Sterling on loan). He had to convert Havertz and Merino into #9’s.

Arteta has a year left on his deal. I’ve talked about his connections with Barca and PSG many times. Despite having a summer final where he finally got what he deserved which saw him win the Premier League title, lose out on the Champions League on penalties in the final, and get to a domestic cup final, Arteta has been left underwhelmed in the market by Arsenal’s board on several occasions.

Mainly good windows (21/22, 22/23, 23/24 and 25/26), sure, and some understandable situations where the club couldn’t invest much because of errors in the past under Emery, but also some windows where the club didn’t go the extra mile, all of which happened all the way up until 25/26 when the club realised Arteta was right and the squad needed two top players for EVERY position.

Plus, throughout that time period, Arteta often had to face internal battles to sign players HE wanted and not guys like Emerson Royal who the club wanted. Make no mistake about it, HE is the guy who understands what Arsenal need in the market more than anybody else. HE is the catalyst behind EVERYTHING good at Arsenal. NOBODY else.

After transforming the club from top to bottom and delivering several seasons which the club have never seen in their entire history DESPITE competing with resources that don’t come close to City, Chelsea and the like, the LEAST an ambitious football club can do at this point is back him significantly with a major signing that he believes elevates the team to continue to win major titles.

Morgan Rogers was that guy.

The club didn’t get it done. That’s the reality. They faffed around all summer and didn’t act with speed or conviction, just like Arteta demanded in his final press conference of the 25/26 season.

Arsenal’s owners are the ones to blame because they pay the fee. Not Berta or anyone else. They are responsible.

@Arsenal, deliver in the market for Mikel OR risk losing him, getting some success despite him leaving in the immediate aftermath because of the foundations that he left, but ultimately regress over a medium-long period of time because he is a once in a lifetime coach that cannot be replaced, like Klopp at Liverpool.

There is still time to make up for it, but Arteta will be furious and rightly so. People talk about the fee, but Arsenal only initiated true contact in the last week. What were they doing for the last month?! They could have negotiated more intentionally… 😤

No wonder there has been silence on Arteta’s contract renewal. Why renew when he has given everything and more but the club cannot give him what he deserves to help the team keep winning over a consistent period of time?

That is why I mentioned PSG and Barcelona earlier. Heck, even Real Madrid or Manchester City. You name the club, they’d want Mikel Arteta in charge.

Respect him or lose him.

Help him evolve the team. To solve the attacking issues of yesteryear with the ‘individuality’ he has claimed the club have needed for many a year.

Part of it is on Arteta to sign the right players and to improve the level by evolving the team with guys like Ødegaard being moved out of the #10 and into a deeper role, for example, but at least give him the tools to be able to do that and evolve the team.

Let’s give the club time to make up for the error of their ways, but this hurts because Rogers was PERFECT for Arsenal…

🙏

[Sanchit Sinha @snsanchit]

@mariyaivasileva I made a post here on this: https://t.co/m2yYOe43kj

[Sanchit Sinha @snsanchit]

One of my #ARR reviewers copy-pasted a Claude-generated review (with line breaks). I could actually recreate the exact text of the review using a generic prompt.
The kicker? They pasted everything EXCEPT the final rating, which was 3.5 on Claude, and they changed it to 2.5.

[Sanchit Sinha @snsanchit]

@DiffKnock I am optimistic, but Fabio Viera and Lokonga also exist in the same universe.

[Sanchit Sinha @snsanchit]

@hllo_wrld Any tips to reconcile with reviewers copy-pasting Claude reviews (which either amplify Limitations or question the experiment settings established in baselines)?
It is so disheartening to see 20 min reviews on a paper with 4 months of work.