Aravind Jayendran

Just closed the lid on the meta-review/AC cycle for ICML 2026. It’s easy to look at a spreadsheet of 4s and 5s and hit "Accept," but the real job is finding the "Reject" papers that actually have merit. I spent today diving into rebuttals and reviews to override committee noise where the science was actually solid. Yes, I accepted papers with reject scores, too. This is where humans still score above AI. Agents can process the "Knowledge" and flag the gaps, but "Judgment", recognizing a nuke in a knife fight, is a uniquely human play. All of our fates (I'm an author, too) are now officially sealed in the cosmic servers of OpenReview. Good luck to all of us.. #icml26 #metareview #research #openreview

@scottdwitt @arnav_kumar Really nice way to put it. These questions may seem annoying or irrelevant at pre-seed, but they do have value.

@arnav_kumar If this CEO IS smart, they'll look up these inconvenient questions, learn the core concepts behind them, and work out solid workable responses to them. (this is one of the ways that I help early-stage founders)

My first-time-founder friend is raising pre-seed for his AI idea. The conversation turned to a certain VC. Let's call this VC Zubin 😀 😀 Zubin is a young VC, looking for his first win. And often rubs people the wrong way by asking overly intellectual but inconsequential questions. Like asking a pre-seed AI company about its projected ebitda margins. Or what happens is facebook decides to get in this category. I asked my friend how the discussion with Zubin went. And he goes - "Let's just say Zubin has the attitude of Peter Thiel, and the talent of Zubin"

Thinking that code generation is going to solve all the issues faced by 100s of millions of current and future business owners is just as accurate as thinking oil refineries are going to solve all of your transportation woes. In fact, the set of challenges/oppty only increases.

Exactly. Code generation commoditises systems knowledge, but what actually solves the business problem is the other kind: failure modes, operational constraints, and domain cycles that only come from running in production. That knowledge compounds in a way no model can shortcut. That gap is widening: aravindjayendran.com/writing/irredu…

AI can replicate your code in weeks. It cannot replicate what your system learned by running. That gap is the moat. Here is what I think is valuable and why It's important to act now: aravindjayendran.com/writing/irredu…

The tech industry convinced an entire generation of developers that they can skip the math. They are wrong. You cannot build foundational architecture with a "GenAI in 5 days" bootcamp. Real engineering requires staring at the equations until they make sense. If you actually want to build state-of-the-art models, you cannot skip the math. No fluff, no quick-bytes. Just the mathematical foundations of Deep Generative Modeling. Here is a course I designed to put everything you need to learn.

Namaste X! This is my first post here - I'm humbled to release my app: VedaVaaNi. It is an interactive learning platform designed to help you practice Rig Veda and Krishna Yajur Veda chanting with absolute phonetic and prosodic (Chandas) precision.

@dahou_yasser This is going to let labs create so much more synthetic data for the next version of foundation models

We are releasing Falcon Perception, an open-vocabulary referring expression segmentation model. Along with it, a 0.3B OCR model that is on par with 3-10x larger competitors. Current systems solve this with complex pipelines (separate encoders, late fusion, matching algorithms). We developed a novel simpler "bitter" approach: one early-fusion Transformer (image + text from first layer) with a shared parameter space, and let scale + training signal do the work. Please check our work ! 📄 Paper: arxiv.org/pdf/2603.27365 💻 Code: github.com/tiiuae/falcon-… 🎮 Playground: vision.falcon.aidrc.tii.ae 🤗 Blogpost: huggingface.co/blog/tiiuae/fa…

@euanashley We recently did a study on LLMs being used in agentic coding and found something similar: blogs.latentforce.ai/blogs/deja-cod…

New AI paper from us this week. When my student first showed me his initial findings, I really didn’t know what to make of them. I felt that this was an interesting but curious loophole phenomenon that would shortly be closed. I was very wrong. arxiv.org/abs/2603.21687

@omarsar0 If you are an ai dev and has not already figured it out ngmi

Anthropic is killing it with these technical posts. If you're an AI dev, stop what you are doing and go read this. It shows, in great detail, how to implement an effective multi-agent research system. Pay attention to these key parts:

We are overwhelmed by the energy at our AI Day in Bangalore! Thank you to our inspiring speakers- Pradyun P Rao, Founder of Alaiy Global Technologies, Aravind Jayendran, Founder & CEO of LatentForce.ai, and Vikas R Jain, Founder, and CEO of Acviss Technologies. This event was organized in collaboration with our strategic partner, Karnataka Digital Economy Mission (KDEM). Their invaluable support has been key to making this gathering of innovators a reality! The speakers explored the cutting edge of AI, diving into: 🔸 How these revolutionary technologies are reshaping the future of AI and pushing the boundaries of innovation. 🔸Transformative insights into multimodal models that integrate text, images, audio, and beyond—revolutionizing how AI works across industries. 🔸Real-time detection, predictive analytics, and the power of AI to fight counterfeiting both online and offline. They highlighted how partnerships are accelerating research and making advanced AI tools available to businesses of all sizes—not just the giants. And, of course, we discussed the importance of responsible AI development to ensure these technologies are used for good. A special shoutout to all the AI enthusiasts who packed the house for AI Day. Until next time! @Maxaravind | @acvissHQ

@thesephist Try Extractors.ai

AI people how are you parsing your PDFs, I need help - TypeScript / Next.js - Both images + text, bonus points for bounding boxes or Markdown - Ideally low latency (<1s for low 100s of pages)

🚀 Exciting News! 🚀 Join us for an inspiring day of AI innovation and insights! On January 18th in Bangalore, we're hosting an AI Day at KDEM, HSR Layout with a stellar lineup of 3 expert speakers: 🔹 Pradyun P Rao, Founder at Alaiy Global Technologies 🔹 Aravind Jayendran, Founder & CEO at LatentForce.ai 🔹 Vikas Jain, Founder & CEO at Acviss Technologies They’ll be sharing their groundbreaking knowledge in the world of artificial intelligence. 🤖✨ 📅 Date: January 18, 2025, 10 AM 📍 Location: KDEM, HSR Layout, Bangalore Don’t miss out on this opportunity to gain actionable insights, explore cutting-edge technologies, and network with industry leaders! 🙌 🔗 Register Now : e2e.ltd/Blr1 @Maxaravind | @acvissHQ #AIDay #ArtificialIntelligence #Innovation #Networking #TechEvent #BangaloreAI #AICommunity #FutureOfTech #AIExperts #EventInBangalore

@AdrianDittmann Don't know why I am listening to this stuff, but X is awesome 😁

@sagarbarvaliya Well, there are a lot of Indians

@karpathy UPI does the job in India

I feel like a large amount of GDP is locked up because it is difficult for person A to very conveniently pay 5 cents to person B. Current high fixed costs per transaction force each of them to be of high enough amounts, which results in business models with purchase bundles, subscriptions, ad-based, etc., instead of simply pay-as-you-go. As an example, I'd like my computer to auto-pay 5 cents to the article/blog that I just read but I can't, and I think we're worse for it. In a capitalist system, transactions between entities are the gradient signal of the economy. Because our pipes don't support low magnitude terms in the sums, the gradients are not flowing properly through the system. I'm not familiar enough with payments to have an idea of specific solutions, but I expect we'd see a lot of positive 2nd / 3rd order effects if the gradients were allowed to flow properly, frictionlessly and with much higher resolution.

@AIMachineDream @JThornbull @fchollet @egrefen @stanfordnlp I think people are underestimating ICL abilities.

The instant training was over the LLM had already figured out how to answer every question it would ever get. Not to say that inference time reasoning can’t be done, just that it isn’t yet. Doesn’t mean that LLMs today don’t reason, but rather they did the reasoning before you asked the question.

I can’t believe I still have to say this, but books are OBVIOUSLY capable of reasoning. You can literally watch them reason IN PLAIN ENGLISH (or whichever language they are written in) in front of your very own eyes….

@corbtt World interactions + continual learning is the way to go🚀

So we’ve got ~6 labs with models at the GPT-4 level, and nobody has meaningfully surpassed it. OpenAI had to scrap/restart their GPT-5 training project bc this barrier has been harder to break than expected. For the next leap a lot of labs seem to be betting on something new. Lotta smart folks pivoting to reinforcement learning (the real thing, not the “RLHF” weak-sauce version). You gotta let your model make contact with the real world, test its ideas, get bruised. The hard part here is making a good general-purpose reward model that isn’t insanely sparse, but maybe not impossible?

@giffmana @joanfihu What if you need to download that file and use it somewhere else? webp doesn't have the level of support that JPEG or PNG has!

@joanfihu All major browsers do

We should never be saving a single jpg or png image anymore. Why are we still doing that? WebP has better compression for same quality, or better quality for same compression. WebP compresses lossless better than png. WebP is supported by all current browsers. Let's WebP!!

@RachelVT42 @orthonormalist @yacineMTB Use Claude to create the UI bruh

@orthonormalist @yacineMTB As long as I can use an LLM in the backend that sounds good to me. I have no shortage of ideas but I don’t have the technical skills to work on UI/UX.

what's the farmville app of LLMs? who's building this?