Tarun Raheja

@satwik1729 @shahkulin98 @mhahn29 Super cool work. Congratulations!

Given black-box access to a Transformer's output, can we efficiently recover its parameters? We analyse the learnability of attention-based models with query access in our new work. Accepted at #ICML2026 🎉 Work done with @shahkulin98, @mhahn29 and Varun Kanade. 🧵

We aim to have <= 25 members of technical staff by the end of the year. Each one of them should own the output and outcome of a sizeable team at another similar organization.

ICL lets models adapt rapidly to changing tasks (✅), but the weights stay frozen - leaving performance gains on the table (⚠️). Fine-tuning (like SFT, RL) reaches a higher perf ceiling (✅), but is slow, can hurt OOD performance, and often reduces plasticity (⚠️). Why not combine the strengths (✅) of both? We introduce Fast-Slow Training (FST): fast weights (prompts) quickly capture task-specific nuances, while slow weights (model parameters) internalize the more general, task-agnostic reasoning patterns that should persist across tasks. FST reaches a higher perf asymptote while being more efficient. Since prompts absorb more of the task-specific information, the parameters do not need to move as much. As a result, the model stays closer to the base model, and preserves more plasticity for learning new tasks!

@_arohan_ skip connections

What are various deep learning architecture modifications you all consider hacks?

Excited to co-found Recursive (@recursive_si) with an exceptional team in London and SF to create AI that experiments on how to safely improve itself, turning compute into knowledge that accumulates in an open-ended process of endless, automated scientific discoveries.

If TIR 2026 has been on your mind, this is the sign to stop waiting. Applications close May 22, and the decisions are being made right now. Before you make yours, join us tomorrow at 6PM IST for the TIR 2026 Pre-Application Webinar. We'll walk you through everything. The program structure, what we look for in applicants, the funding, the ecosystem, and what life at ARTPARK actually looks like. Ask us anything. We're answering it all, live. This is your chance to get every question answered before you hit submit. Don't let May 22 catch you off guard. 🔗 Register now: us05web.zoom.us/meeting/regist…

@DrDatta_AIIMS @vinayprabhu Pune, yep!

@vinayprabhu @atemyipod Tarun is in Mumbai?

Going for a game theory workshop to learn from the best in IIT Bombay! If you are a researcher in healthcare AI or want to upgrade your hospital with AI, let’s catch up and exchange notes. Have a series of meetings in Mumbai and Thane. Can consider coming to Pune if there is enough interest! DMs open!

For years, the best minds from India's top institutions have taken their research to the world, and they've done remarkable things there. But something is shifting. The problems worth solving are here. The research talent is world-class. The infrastructure is catching up. At ARTPARK- IISc Bengaluru, we're building the place where deep-tech research becomes real products with the funding, mentorship, and ecosystem it deserves. That's what the Technology Innovators in Residency is. 12 months at ARTPARK, up to ₹1 Crore in funding, direct access to IISc faculty, industry leaders, and a network of founders who've been exactly where you are. No company required. Just your research, your vision, and the drive to build. If you've been sitting on an idea that you know is ready for the next step, this is worth a look. 𝘓𝘦𝘵’𝘴 𝘯𝘰𝘵 𝘫𝘶𝘴𝘵 𝘮𝘢𝘬𝘦 𝘪𝘯 𝘐𝘯𝘥𝘪𝘢, 𝘣𝘶𝘵 𝘣𝘶𝘪𝘭𝘥 𝘪𝘯 𝘐𝘯𝘥𝘪𝘢. Applications close May 22, 2026. Apply here: apply.artpark.info

𝐀𝐑𝐓𝐏𝐀𝐑𝐊 𝐢𝐬 𝐡𝐨𝐬𝐭𝐢𝐧𝐠 𝐚𝐧 𝐎𝐩𝐞𝐧 𝐃𝐚𝐲 𝐨𝐧 𝐌𝐚𝐲 14, 2026. Come meet the startups and innovators building in AI, robotics, and advanced technologies at our ARTgarage facility. Get a closer look at how ideas move from lab to real-world deployment and see the work happening on the ground. 10:30 AM to 1:00 PM | ARTGarage, Jalahalli, Bengaluru Register now to reserve your spot: docs.google.com/forms/d/e/1FAI… Please note that the confirmed registrations will receive a follow-up email with details. #ARTPARK #OpenDay #DeepTech #Robotics #Bengaluru #Innovation

“The fastest way to shorten the robot gap is not better demos. It is harder, more honest engineering against real customer ROI, and the patience to let the deployment flywheel compound before declaring victory.” 👊

Still figuring out if TIR is the right fit for you? Join us tomorrow at 6PM IST for the first pre-application webinar. We will walk you through the program, answer your questions, and help you decide if this is your moment. Register now: lnkd.in/gk2BpCfd Applications close May 22. #TIR #ARTPARK #IISc #DeepTech #Bengaluru #OpenForApplications

If you have a computational PhD and are interested in Biology/Healthcare or in healthcare (at least MD and 1 year of computational research experience) and interested in becoming fully computational, I highly recommend the Simons-Ashoka Early Career Fellowship for you. India needs more computational healthcare personnel urgently. I am looking at Radiology. Mayank looks at ICU. We are both Simons-Ashoka fellows here. At this point of time, we urgently need people who are interested in computational pathology, dermatology, NLP in healthcare, etc. If you have a PhD or are an MD/MS and want to get into computational medicine in healthcare full-time, you should definitely consider joining us! If you fit the eligibility criteria, feel free to DM!

𝐓𝐡𝐞 2026 𝐀𝐑𝐓𝐏𝐀𝐑𝐊 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐲 𝐈𝐧𝐧𝐨𝐯𝐚𝐭𝐨𝐫𝐬 𝐢𝐧 𝐑𝐞𝐬𝐢𝐝𝐞𝐧𝐜𝐞 𝐩𝐫𝐨𝐠𝐫𝐚𝐦 𝐢𝐬 𝐧𝐨𝐰 𝐨𝐩𝐞𝐧. If you have lab-validated deep-tech IP and the ambition to build something real, this is your shot. 12 months at ARTPARK-IISc Bengaluru, funding, infrastructure, and a clear path from research to product. Have questions? Join our pre-application webinars on May 6, 13, and 18 designed specifically to walk you through the process and answer everything you need to know before applying. Swipe through for the full details. Apply by May 22, 2026. Link: apply.artpark.info #deeptech #research #product #applynow #funding #labtomarket #residenceprogram

𝐓𝐡𝐞 2026 𝐀𝐑𝐓𝐏𝐀𝐑𝐊 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐲 𝐈𝐧𝐧𝐨𝐯𝐚𝐭𝐨𝐫𝐬 𝐢𝐧 𝐑𝐞𝐬𝐢𝐝𝐞𝐧𝐜𝐞 𝐩𝐫𝐨𝐠𝐫𝐚𝐦 𝐢𝐬 𝐧𝐨𝐰 𝐨𝐩𝐞𝐧. If you have lab-validated deep-tech IP and the ambition to build something real, this is your shot. 12 months at IISc Bengaluru, funding, infrastructure, and a clear path from research to product. Have questions? Join our pre-application webinars on May 6, 13, and 18 designed specifically to walk you through the process and answer everything you need to know before applying. Swipe through for the full details. Apply by May 22, 2026. Link: apply.artpark.info

@sahilgo6801 @GoogleDeepMind @jainprateek_ @adityakusupati @Swayam_421 @paul_sujoy_ @ManishGuptaMG1 @gautham_ga_ Super cool. Congratulations!

🧵[1/7] Excited to share our new research from @GoogleDeepMind ELT: Elastic Looped Transformers for Visual Generation Introducing a highly parameter-efficient class of visual generative models 1. High Throughput 2. Elastic/Any-Time Inference

Want to work at the absolute frontier of robotics? We're hiring for a founding researcher to wrap up our founding team. 📌Full time, NCR Based 🔥50 LPA + significant equity ✅Application link in the comments.

@itsalexvacca Repo

We’re building out our embedded engineering team to develop the next generation of data collection hardware for robotics. Multiple roles- all based in Delhi NCR. 25-30 LPA CTC. If you can debug something you cannot see, please apply. Link in comments. Rt for good karma 🙏

@matt_gray_ INSTALL

This is the biggest announcement I’ve ever made in my entire business career. For the last 7 years, I’ve built multiple personal brands to multi 7 & even 8-figures. Today I can officially say my team can install the EXACT same system into your business to help you generate more leads, visibility, and authority across ALL platforms. If you’re serious about growing this year and know that a personal brand is the #1 lever to get there, Comment INSTALL and I’ll reach out directly (follow me first or I can't DM you)

@yanndine Gtm

A two-person GTM team at a Series B SaaS company closed $2.4M in pipeline in one quarter. No SDRs. No demand gen agency. No paid ads. Signal-based outreach. Intent scoring. AI-sequenced follow-up. Automated reporting. Two GTM engineers running the whole motion - for one quarter. I pulled it apart. Compared it to every system we've built across the GTM teams we've worked with. Then asked myself one question: If I had to reverse engineer this from scratch - what would it actually look like? Turns out the architecture isn't that complicated. I mapped the whole thing into a step-by-step playbook you can upload directly to any LLM. It walks you through building your own version from GTM strategy to fully AI-powered execution. Comment "GTM" and I'll send it over.