Written a blog on why LLM Hallucination might be a feature then a bug. This is based on the paper arxiv.org/abs/2409.05746 . Here is the blog medium.com/@jain.sm/why-l…

I just published What Has a Foundation Model Actually Learned?Does it learn the world model or not? medium.com/p/what-has-a-f…

I just published What Mixture-of-Recursions Actually Does medium.com/p/what-mixture…

I just published How the Brain Learns Concepts: A Grounded Architecture of Memory, Reward, and Simulation medium.com/p/how-the-brai…

I just published EvolDSL: Neural-Guided Self-Bootstrapping Programming Systems medium.com/p/evoldsl-neur…

experimented with notebook LM video preview for a video on Bayesian brain. Some rough edges IMO , but still really really good

Gemini allows us to create short story books and its really good..Here is a small one I created gemini.google.com/share/138decb2…

I just published Building Self-Improving AI Agents with Bayesian Meta-Learning medium.com/p/building-sel…

Trained a small model to play flappy bird and chess using PPO /RL. The goal was not to get the best chess engine but see how it can be done. The code is here github.com/allthingssecur…

48 strotas of Jain text Bhaktambara youtu.be/ZU-RHWdFdo0?si… Rohit Jain 🇮🇳 Ocean Jain

I just published Beyond brittle facts: building a probabilistic knowledge graph medium.com/p/beyond-britt…

AI Agents aren’t just software routines. They’re inching closer to how our own minds work—memory, attention, decision loops, even self-correction. I’m putting my thoughts into a living book—drawing from cognitive neuroscience to design agents that think and adapt more like us .

It took me around 30 mins to collaborate with Claude code and generate an entirely new programming language "Bhagwad", based on preachings and principles of Bhagwad Geeta. The language uses constructs like types, syntax from Bhagwad Geeta and uses a custom lexer, parser, and

🚀 Breakthrough in AI-assisted mathematics! Math, Inc.'s Gauss autoformalization agent has completed the strong Prime Number Theorem formalization in just 3 weeks - a project that challenged Fields Medallist Terence Tao for 18 months. Key achievements: • Automated complex

What I understand is that we can define intelligence as a combination of 4 major principles. We compress the world into smaller patterns to build meaningful abstractions. Search through the options in front of us (meaning avoiding brute force) Predict what will happen next,

Neural nets are amazing pattern matching engines. But pattern-matching alone isn’t the end of the story. Human knowledge has always advanced through explanations, ideas that are hard to vary, that survive criticism, and that let us imagine what would happen if things were

Sora is just an amazing model..

The quantum universe is like all possible stations broadcasting at once. A rock has no tuner . it just sits in the static. A brain is a dynamic tuner that locks onto one frequency pattern, sustaining it over time. That “locked-on signal” is what we call experience.

Using codex built a web app for having Radio feeds from across the globe. allthingssecurity.github.io/globalfreq-ai/

It was a mindboggling experience letting Claude opus create this awesome game for me. The game is based on one of the great works of Richard Dawkins The Blind Watchmaker..Works on desktop for now as my Opus limit exhausted to fix it for mobile. Here is the link to experiment and