Nevasini

what's in every session, all synced: – RGB @ 15fps – LiDAR depth per frame – ARKit 6-DoF camera pose – IMU @ ~100Hz – 21-joint MANO two-hand mocap – room mesh – hierarchical language (session → sub-goal → episode → atomic action)

Been digging through STERA-10M all morning. some notes: 10M FPV samples for training world models, just released by @fpv_labs on Hugging Face. a few reasons I think this matters for embodied AI / spatial intelligence 🧵 huggingface.co/datasets/fpvla…

most egocentric datasets are 30-second clips or short capped sessions. stera goes the other way: 20.5 min mean, 104 min longest continuous. if you're doing VLA pretraining or long-horizon manipulation, this gap matters a lot.

first, the part that's actually wild: it was all captured on a regular iPhone Pro. no Project Aria, no custom rig — just a head-mounted phone running their open capture app. 200 hours. 584 sessions. 20 contributors. 20+ environments.

@radbackwards Highly interested!

NEO is in NYC right now. Doing a meet and greet Saturday night if you want to say hi to NEO & the team— lmk. Limited availabilities.

@shashin_bhaskar Woah🥺

> came to SF > bored > looked up on going events > went to Google Hackathon > banging food and won the hackathon

📝 Introducing "𝐀𝐥𝐥 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬 𝐌𝐚𝐭𝐭𝐞𝐫: 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐧𝐠 𝐋𝐌𝐌𝐬 𝐨𝐧 𝐂𝐮𝐥𝐭𝐮𝐫𝐚𝐥𝐥𝐲 𝐃𝐢𝐯𝐞𝐫𝐬𝐞 100 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬 (𝐀𝐋𝐌-𝐁𝐞𝐧𝐜𝐡)" ➡️ 𝐃𝐢𝐯𝐞𝐫𝐬𝐞 𝐌𝐮𝐥𝐭𝐢𝐥𝐢𝐧𝐠𝐮𝐚𝐥 𝐁𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤: A culturally diverse multilingual and multimodal VQA benchmark covering 100 languages with 22.7K question-answers. ➡️ 𝐂𝐨𝐦𝐩𝐫𝐞𝐡𝐞𝐧𝐬𝐢𝐯𝐞 𝐃𝐨𝐦𝐚𝐢𝐧 𝐂𝐨𝐯𝐞𝐫𝐚𝐠𝐞: ALM-bench encompasses 19 generic and culture-specific domains for each language, enriched with four diverse question types. ➡️ 𝐄𝐱𝐩𝐞𝐫𝐭-𝐕𝐞𝐫𝐢𝐟𝐢𝐞𝐝 𝐀𝐧𝐧𝐨𝐭𝐚𝐭𝐢𝐨𝐧𝐬: With over 800 hours of human annotations, ALM-Bench is meticulously curated and verified with native-language experts to assess the next generation of massively multilingual multimodal models in a standardized way, pushing the boundaries of LMMs towards better cultural understanding and inclusivity. 🔹 "𝐀𝐥𝐥 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬 𝐌𝐚𝐭𝐭𝐞𝐫: 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐧𝐠 𝐋𝐌𝐌𝐬 𝐨𝐧 𝐂𝐮𝐥𝐭𝐮𝐫𝐚𝐥𝐥𝐲 𝐃𝐢𝐯𝐞𝐫𝐬𝐞 100 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞𝐬 (𝐀𝐋𝐌-𝐁𝐞𝐧𝐜𝐡)" 🔹 In collaboration with @MBZUAI, @UCF, @AaltoUniversity, @our_ANU, and @liu_universitet 📝 Paper: arxiv.org/pdf/2411.16508 🔗 Project: mbzuai-oryx.github.io/ALM-Bench/ 🤗 Dataset: huggingface.co/datasets/MBZUA… 💻 Code: github.com/mbzuai-oryx/AL… 🙏 Thanks to the entire core team and all the volunteers. #ArtificialIntelligence #GenAI #research

Hot take: one device one pair earphones different music on each ear 🥺

Just AI fantasies😉

Hot take: Slack displays dm's based on whom you texted first rather than recent

Everything is LLM'ing😑

OpenAI or Google now? 🤔

This woman beside me has blocked focus time on her office teams calendar 🥺🥺

@vegfriedmomos @aditeyabaral01 so they are friends who are not friending?🤭

@aditeyabaral01 sir please understand the bhavana

first official solo trip for conf with friends and my periods started. the W never comes without the L

Switching the GPU on and off seems like a typical watchman 😂

carrying 3 laptops to office, what a mess can I be😑

@BalajiAI Really have to try this🥲

This is horrible😱

WhatsApp should really have auto translate @AIatMeta @Meta

@i_sudeepa nevasini24@gmail.com

I've compiled guides, samples, and walkthroughs for Statements of Purpose (SoP), and Letters of Recommendation (LOR) from top US universities. If you're interested, comment your email, and I'll send it your way.