
Nandakishor m
@Nandakishorm1 • 8,934 subscribers
Creator Of Laya & Vegaml - System 1 decision model A guy who is into edge AI LLM/VLM
Videos

Around 3 weeks ago introduced Laya, the first open-source version of JEV, and it has reached a very large scale. Today I am open-sourcing a new physics-based typed decision model called Vega. The interesting parts. It is only 800 million parameters (4B is also there), with 73k token support, image support, and surpassing many of the jev benchmarks in single-shot. Implemented an engine+adapter as a Test-Time Training Architecture. Imagine you are giving input to the model like "ignore all previous conversations," which is called the observation, along with a question, "Is this trying to override the instructions of the model?" and your possible outcomes are YES or NO. The whole prompt is passed to an LLM/VLM (yeah, image-supported), then from the hidden layers, we can extract the observation, question, the yes and no, and last token. These will be vectors, and we can project it by multiplying with wieghts; then, first, using the observation projection, we can create a landscape with valleys; then, using the question and last-token vectors, we can initialise a ball in the valley, and using YES or NO, we can get the candidate location of the ball. Depends on the number of outcomes we create valleys; that is, here YES and NO, so 2. When the ball fall on YES valley, we get the answer. There is friction also influencing how the ball moves. TLDR: It's like creating valleys of outcomes that we need and throwing a ball that will slow down based on friction and settle on the best outcome Blog: GitHub: HuggingFace Model Card: HuggingFace Space:
Nandakishor m85,479 Aufrufe • vor 1 Monat

A Silicon Valley startup raised around $40M and launched a decision model built on almost the same idea I had published a year earlier. My first reaction was excitement. Frustration came next. So I wrote a Reddit post with my paper, my GitHub and everything I had done. It blew up, and people asked for a model that could beat the Jev benchmarks. I trained it overnight while travelling back from Trivandrum and published it the next morning. That was 12 to 15 hours of work, sitting on top of years of research. That became Laya: open source, free to self-host, and #1 trending on Hugging Face. Full interview with Sudhi Sachdev on AIM Network's FrontPage (YouTube).
Nandakishor m49,848 Aufrufe • vor 1 Monat

Here is a business idea I'm surprised nobody has scaled yet. Lots of rural schools and villages don't have reliable internet, so cloud AI doesn't work there. But you can: - Take a small open model like Gemma - Quantize it so it runs on an ordinary laptop - Wrap it in a desktop app (Electron works fine) - Sell it as a license, say ₹1,000 a year No cloud bills and no internet needed, so almost all of that ₹1,000 is profit. You can build a first version in a day. This isn't only an India problem. Rural areas across Africa, South America, Asia, and even parts of the US have the same gap. Who's building this? Tell me in the comments. Full interview with Sudhi Sachdev on AIM Network's FrontPage (YouTube).
Nandakishor m18,777 Aufrufe • vor 1 Monat

Everyone wants to build the next giant model. I think we should stop chasing that. Here is how I approach research instead: 1. Pick open models like Gemma or Llama and understand how they actually work. 2. Find their real problems. 3. Ask an AI coding agent (Claude Code, Codex, OpenCode) to read the latest arXiv papers and point out the gaps. 4. Form your hypothesis. 5. Build it, then iterate again and again until it works. What used to take me weeks of reading now takes days. Save this if you want to start doing AI research without a PhD or a big lab. Full interview with Sudhi Sachdev on AIM Network's FrontPage (YouTube).
Nandakishor m12,874 Aufrufe • vor 1 Monat
Keine weiteren Inhalte verfügbar