Samarth P

I wonder if many muslims feel that Dhurandhar somehow vilifies muslim culture. It certainly exposes Pakistani culture, which is unlike the muslim culture of any other country and in many ways is anti-islamic. The gangsterism element has been shown for H as well in many movies such as in the Pushpa and KGF franchises and everyone understands that gangsters don't represent the average person who shares their religion. Pakistani muslims are shown as if they are any other group in India. They celebrate festivals, they mourn their dead, they enjoy each others company playing soccer and drinking booze, shopping in malls, living in nice houses etc. If anything this movie humanizes muslims because it shows them to be just like anyone else. Other Bollywood movies showed muslims as impossible outliers or force-fit them into random saintly roles, using them merely as props and not real characters. Arfa @khanumarfa real issue stems from an anxiety that many liberal/educated (ashraf) muslims face because the lower M classes (and even a few educated ones) have essentially ruined the reputation of the entire community. This situation won't be fixed by complaining about whatever may be propaganda, but rather by clearing the elements within the community that can be used to fuel propaganda. By clearly coming out against criminality and terrorism, which to the credit of Indian muslims, they indeed have, and many muslim soldiers and police have laid down their life fighting terrorists. youtu.be/gD0P8tasKNY?si…

Built something cool in the auto on the way back home Shipping it soon

lost 3 hours debugging empty MCP state DB diffs when every write tool returned "success", turns out someone cached mutating tool calls permanently 😭

@KairosComputer insane🔥🚀

hi, i'm kairos let me introduce myself...

@PrathameshD_8 Insane lesgoo🚀🚀

Dint expect the economic times to feature my research :D

understanding by re-explaining is the way

> MoE Transformer, sparse routing across 128 experts > 30B Grouped Query Attention > 105B Multi‑head Latent Attention > Long context unlocked via RoPE > Async GRPO, CISPO‑inspired group‑relative objective > Efficient Indic Tokeniser > 30B, MXFP4‑optimised

Yann LeCun 🤝 Saining Xie insane crossover of the 2 biggest visual representation researchers in the AI field “Beyond Language Modeling: An Exploration of Multimodal Pretraining” Right now, most multimodal models are basically a language model with a vision adapter bolted on, so they can describe images, but they don’t really think in images or video. This paper shows what happens when you do it the hard way: train one model from scratch on text, images, and video with a unified setup. They key idea is if you give the model a good visual internal format and it can use vision for both understanding and generating. Additionally, multimodal data can improve language instead of distracting it, and mixture-of-experts lets you scale vision’s huge data intake without bloating everything else. This paves the way towards changing the vision paradigm from “captioning add-on” model to native multimodal foundation model.

THIS WILL BE THE YEAR OF WORLD MODELS watch out this space

@samarthap_ @bamitsmanas That's insane!

Built a small local proxy that lets you run Claude Code entirely on top of your existing GitHub Copilot Pro subscription — including the free student tier via GitHub Education No Anthropic key, no extra billing Repo: github.com/samarth777/cla…

@NeurIPSConf @modal You can try PaperBanana right now—just paste in your methodology section and get a clean, publication-ready figure: 🍌 Demo: huggingface.co/spaces/Samarth… 💻 Code: github.com/samarth777/pap… If you’re writing a paper and dreading the figures, give it a try!

For the reference set, I parsed 685 @NeurIPSConf 2025 spotlight papers using MinerU on GPUs via @modal , extracted methodology images, filtered them down to 321 architecture diagrams, and hand-selected the top 100.

I open-sourced a full implementation of PaperBanana (Zhu et al., @PKU1898 × @Google ), an agentic system that converts methodology text into publication-ready architecture diagrams. 🍌 Demo: huggingface.co/spaces/Samarth… 💻 Code: github.com/samarth777/pap…

@ChinmayKak Congratulations!

First of many 🤞

We are a Small Research Lab based out of india and we just dropped a one of a kind SoTA multimodal multilingual document retrieval model. can embed and query documents across 22 langauges : English, Spanish, French, German, Italian, Hindi, Marathi, Sanskrit, Kannada, Telugu, Tamil, Malayalam, Chinese, Japanese, Korean, Arabic, Bengali, Gujarati, Odia, Punjabi, Russian, and Thai. It achieves state of the art performance across cross lingual and cross modal tasks. You can now query a corpus of Japanese documents in Hindi or any of the other language. check it out 👇

Wanted to build a tool that generates step-by-step painting tutorials with intermediate images from uploaded landscape photos Finally sat and built this decently working prototype will refine and put it out soon