Swami Chandrasekaran

This is what I did this summer with 9-10 year old girls. Teaching them hands-on a variety of key topics incl #Python #DataAnalysis #PowerBI and many more to follow. #STEM #WomenWhoCode

Agent = Model + Harness. Fine. So what’s the agent harness? So I built dadloop—an executable agent harness explained through a suburban dad (with Mom as the governance layer). Substack: nirvacana.substack.com/p/the-part-eve… GitHub: github.com/swamichandra/d…

@GoogleAIStudio LLM MRI. Understand the planning, thinking, and deficiencies.

What are you vibe coding this weekend?

Somewhere in your enterprise right now an agent harness is summarizing its own summaries at 2am. The hogger is asleep. Proud of the overnight run. That’s not transformation. It’s tokenmaxxing. Mirror test inside. #TokenDuck #WhoEarnsOpus # EnterpriseAI @nirvacana/note/p-196845299?utm_source=notes-share-action&r=5qmw" target="_blank" rel="nofollow noopener">substack.com/@nirvacana/not…

Everyone keeps bragging about their models and agents. Congrats — you built employees with no office. The Work Surface is the layer nobody built, the one that makes whole processes vanish and outcomes run themselves. Explained here: . open.substack.com/pub/nirvacana/…

Built a spec-driven AI Skill: “Ship-it-Ralph”. Give it an idea in one sentence → working app out. Runs the Ralph loop: spec → tasks → tests → code. Use in GH Copilot, Cursor, Claude Code, or Antigravity. github.com/swamichandra/s… #specdrivendevelopment #ralphloop #skill

@mattpocockuk Are you using this outside of Claude Code or Claude Desktop? I’m very curious to use the Skills outside of the Anthropic ecosystem of tools esp the execution of scripts.

Skills I'm currently running: - write-a-prd - make-refactor-request - tdd - design-an-interface - write-a-skill

What if there was a Snakes & Ladders for AI Agents? Agentic Systems Ladder—81 cells mapping the agentic systems journey from "should this exist?" to sustained value. The snakes are waiting. Are you ready? #AIAgents #EnterpriseAI #AgenticSystems #AI2026 nirvacana.com/thoughts/2026/…

@grok @newstart_2024 What are some of the top-15 household items incl ones for men that are considered toxic per the Think Dirty app?

Based on Think Dirty app data and user reports, here are 10 examples of products often rated highly toxic (8-10/10): 1. Barbasol Shaving Cream - 9/10 2. St. Tropez Self Tan Mousse - 9/10 3. Jergens Natural Glow Moisturizer - 9/10 4. Moroccanoil Dry Shampoo - 9/10 5. Donna Karan Cashmere Mist Deodorant - 9/10 6. Urban Decay All Nighter Setting Spray - 8/10 7. Bobbi Brown Long-Wear Gel Eyeliner - 9/10 8. It's a 10 Miracle Leave-In - 9/10 9. Laura Mercier Translucent Powder - 8/10 10. Living Proof PHD Shampoo - 9/10 Scan products yourself for latest ratings.

Dr. Daniel Amen's top daily habits to grow & protect your brain (at any age) - Exercise (coordination + strength) boosts blood flow - Learn new things — retirement is when brains shrink fastest - Get blood work yearly (check ferritin — high iron ages you; donate blood 2x/year) - Omega-3s (fish oil or healthy fish) = more gray matter - Curcumin (turmeric) fights inflammation & depression - Saffron for mood & brain support - Avoid toxins (scan products with Think Dirty app — e.g., Barbasol = 9/10 toxic, switch to safer) Every day: Choose brain-building over brain-shrinking. 3:49 clip inside — science-backed, life-extending routine.

KPMG US's Q1 #AI Pulse Survey 2025 shows a GenAI investment surge. Leaders plan to boost GenAI investment to $114M next year. AI-agent pilots are up from 35% to 65%, with 82% of leaders citing risk management as the top challenge. ow.ly/sgtn50VBEnv

@svembu Sorry for the shameless plug. Recently wrote about on this topic. linkedin.com/pulse/hallucin…

All the LLMs and other deep learning models are based on neural networks. We can think of them as mathematical functions with hundreds of billions of parameters. Those parameters (weights) are determined during training and we train these networks with trillions of tokens (text, images, videos that are split up into tokens to be ingested by the models). We can say that every one of the trillions of tokens played a part in determining the value of each of the hundreds of billions of parameters. The image I have in mind is a giant lake where we dissolve trillions of cubes of salt, sugar etc. After the dissolution we cannot know which of the cubes of sugar went where in the lake - every cube of sugar is everywhere! Therein lies a problem: if we use a business database, such as customer relationship data, to train a neural network model (i.e to determine its parameters), when the customer changes that data or deletes the data, we do not know how to alter the weights of the model to account for this change in the data. Even if the model were dedicated to that customer, we still cannot guarantee the customer that their changes to the data will be reflected in the model. In that sense, neural networks (and therefore LLMs) are NOT a suitable database. This is a fundamental limitation of the current scientific mathematical approach and cannot be fixed only by technological fine tuning. The RAG (retrieval augmented generation) architecture keeps the business database separate and augments the user prompt with data fetched from the database. In that case, the model itself is not trained on the (potentially changing) customer data because that data is only used in the prompt. But RAGs can only go so far. I personally have come to believe more foundational work is needed. What does that look like? All I have right now are hunches. That is the existing part of scientific work!

@DrJimFan I’m curious as well. Can one take any of the larger parameter models and force it to self-reflect and thereby forcing increased TTC? Could be via a wrapper meta prompt? Or this emergent behavior is observed only when you start to add verifiers, best of N, beam search, PRM etc.

We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all. It makes no sense. The most entertaining outcome is the most likely. DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets. They are perhaps the first OSS project that shows major, sustained growth of an RL flywheel. Impact can be done by "ASI achieved internally" or mythical names like "Project Strawberry". Impact can also be done by simply dumping the raw algorithms and matplotlib learning curves. I'm reading the paper: > Purely driven by RL, no SFT at all ("cold start"). Reminiscent of AlphaZero - master Go, Shogi, and Chess from scratch, without imitating human grandmaster moves first. This is the most significant takeaway from the paper. > Use groundtruth rewards computed by hardcoded rules. Avoid any learned reward models that RL can easily hack against. > Thinking time of the model steadily increases as training proceeds - this is not pre-programmed, but an emergent property! > Emergence of self-reflection and exploration behaviors. > GRPO instead of PPO: it removes the critic net from PPO and uses the average reward of multiple samples instead. Simple method to reduce memory use. Note that GRPO was also invented by DeepSeek in Feb 2024 ... what a cracked team.

How Far Can LLMs Reason? Dive into a prompt of epic reasoning—pun intended—featuring LLMs, the Mahabharata’s Chakravyuha, and Abhimanyu’s ultimate challenge. linkedin.com/pulse/epic-rea…

Check out my latest article: Houston: We have a Path-to-Production Problem in AI linkedin.com/pulse/houston-… via @LinkedIn

@JonathanRoss321 From a system architecture perspective (IEEE 42010): - Prompts are 'components’ - LLM is a 'system element' - Components (prompts, orchestration, etc.) & system elements (LLM, UI, observability, etc.) form the 'Prompt-based System-of-Interest'.

When we build something out of hardware we call it a System. When we build something with software we call it an App or Application. What should we call the thing we build using prompts?

21. Metro (1982) Composer: #Ilaiyaraaja J. Mahendran got the classy best, consistently. S. Janaki’s humming in “Metti Oli Kaatrodu" and “ Santhakavigal” - one of a kind. youtu.be/QWX-cn6dIhM?si…

20. Mantri Gari Viyyankudu Composer: #Ilaiyaraaja I discovered this gem late. This album stands as one of #Chiranjeevi career bests. Sample “Chi Chipa” to understand why. Cracker of an album that packs a punch! youtu.be/qS5QA-B2qEE?si…

Here is my "60 Top Indian Film Music Albums of All Time" list. Albums that broke new ground, *every* song exhibited raw brilliance & genius of the composer. Tough one to do. If you don't find your favorite(s) here, incl nursery rhymes, pls create your own list! #IFM60 Thread⬇️