Parth Bhalerao

[Parth Bhalerao @ParthVBhalerao]

Excited to share that When Cultures Move has been selected at EMNLP 2026 for the NLP4PI Workshop.

We study multicultural video generation, building on our earlier ACL 2026 work on multicultural image generation.

Thanks to @OanaIgnatRo for the guidance.

https://t.co/8dBh1LjLFG

[thamar | @thamar_solorio]

If we want to stop the crazy rush for the next paper, we need to stop counting when papers get accepted. We should go back to focusing on the science, what is the paper about? Why we would like to read it? Not really whether you got x/y papers in.
We're loosing sight of what truly matters...
#NLProc #publishorperish #beancounting

[Parth Bhalerao @ParthVBhalerao]

@adityapuranik99 @OanaIgnatRo Thank you ! 😊

[Parth Bhalerao @ParthVBhalerao]

Excited to share that our paper “Beyond Factual QA: Mentorship Oriented Question Answering over Long Form Multilingual Content” has been accepted to EMNLP 2026. (Findings)

https://t.co/E4VK0EUErI

Very grateful to @OanaIgnatRo for her guidance and support throughout this work.

[Parth Bhalerao @ParthVBhalerao]

GRADE, our work on evaluating AI tutors in student-tutor dialogues!

Open-source LoRA pipelines matching proprietary systems + carbon footprint analysis included 🌱

Thanks to my Advisor Dr. @OanaIgnatRo !

Code: https://t.co/iFUKl3AdXZ
Preprint: https://t.co/GNfsZYyH7p

[Parth Bhalerao @ParthVBhalerao]

My first paper accepted to ACL Findings 2026! When Cultures Meet: Multicultural Text-to-Image Generation w/ advisor @OanaIgnatRo 9K multicultural images — 11k+ downloads!

🤗 Dataset: https://t.co/7jDyTS0OF7
💻 GitHub: https://t.co/nBsTrikEVl
📖 Paper: https://t.co/ewM3xIvKa0

[primordial intelligence @primal_brainer]

The older you get, the more you realize luck is mostly exposure.

If you sit in the same place, have the same routine, talking to the same people, nothing new really happens.

You have to tackle the world to win.

Travel more. Talk to people. Try a breakfast spot. Post on social media. Start a side hustle or a hobby.

The world rewards motion. You don't find opportunity sitting still.

[DAN KOE @thedankoe]

You need to write more.

Without AI. Without templates. Without knowing what you're writing about. Just you, an idea, and enough time to do the difficult cognitive work necessary to reach true understanding. If you don't, your ability to think will drastically decline.

[Parth Bhalerao @ParthVBhalerao]

Research keeps reminding me: progress isn’t just breakthroughs — it’s patience, careful verification, and quiet persistence. Really humbled by this phase of my PhD journey

[HyoJung Han @h__j___han]

Lots of work on cross-lingual alignment encourages multilingual LLMs to generalize knowledge across languages.

But this push for uniformity creates a tension: what happens to knowledge that should remain local?

We look into this trade-off of transfer and cultural erasure:🧵

[Parth Bhalerao @ParthVBhalerao]

@rumilyrics Flow

[Andrej Karpathy @karpathy]

Excited to release new repo: nanochat!
(it's among the most unhinged I've written).

Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI.

It weighs ~8,000 lines of imo quite clean code to:

- Train the tokenizer using a new Rust implementation
- Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics
- Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use.
- SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval)
- RL the model optionally on GSM8K with "GRPO"
- Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI.
- Write a single markdown report card, summarizing and gamifying the whole thing.

Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc.

My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved.

Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

[Celestara @celestaracosmos]

"Beautiful days do not come to you. You must walk towards them." - Rumi

[Wilka Carvalho @cogscikid]

A limitation of LLMs: tacit knowledge---the knowledge you can't put into words

LLMs have all the world's written knowledge. But many important things were never written down.

How do you teach ice skating through text? "Bend your knees, shift forward, arms out for balance." But what about the feel of the blade catching ice? The micro-adjustments your body makes? The sensation of momentum?

Same with dancing - the rhythm in your body, reading your partner's movements. Or coffee tasting - not just "bitter" or "smooth," but the complex flavors that build a sommelier's expertise.

Even in conversation, we constantly use knowledge we can't explain. Sensing emotional undercurrents in someone's voice. Knowing when to pause. Feeling when someone needs encouragement versus honesty.

Right now, LLMs have little to no access to this knowledge

So when we need to incorporate this knowledge into decision-making, there really is no replacing humans

[Jessy Lin @realJessyLin]

🔍 How do we teach an LLM to 𝘮𝘢𝘴𝘵𝘦𝘳 a body of knowledge?

In new work with @AIatMeta, we propose Active Reading 📙: a way for models to teach themselves new things by self-studying their training data. Results:

* 𝟔𝟔% on SimpleQA w/ an 8B model by studying the wikipedia docs (+𝟑𝟏𝟑% vs plain finetuning)
* a domain-specific expert model: 𝟏𝟔𝟎% vs FT on FinanceBench knowledge
* an 8B wikipedia expert competitive w/ 405B on factuality (💥open-sourced!)

🧵[1/n]

[Parth Bhalerao @ParthVBhalerao]

@RishiBommasani Hi rishi,
Looking forward to collaborate.

I have applied !

[Samir Bakre | समीर बाकरे @bakre_samir]

#Motivation #Inspiration #Quotes #PositiveVibes #PositiveThinking #DailyQuotes #SuccessMindset #InspirationalQuotes #MotivationalQuotes #GoodVibesOnly #SelfImprovement #SBThoughts

[Rada Mihalcea @radamihalcea]

🐕 🐒 It was a special treat to give a keynote today on one of my favorite topics: AI for animal communication. This was my first time giving a joint keynote, alongside my brilliant students Artem Abzaliev & @AlvaroVHD. Many thanks to the organizers of the INAOE Symposium on Computational Technologies for Animal Behavior Analysis for the invitation!

Main takeaways from our talk:
1️⃣ This is a highly impactful area of research, with the potential to open doors to understanding other forms of intelligence and to drive advances in conservation, biodiversity, and animal welfare.
2️⃣ It is also a very challenging field: the standard AI pipeline does not apply in a straightforward way, and there are millions of species to account for.
3️⃣ Cross-species AI models show promise as a solution, and we presented concrete steps toward decoding the vocalizations of dogs and capuchin monkeys.

[Verena Rieser @verena_rieser]

Absolute star panel ⭐ at @NLP4PosImpact