←
Bargava9,580 posts
Bargava avatar

Bargava

@bargava

I build AI and AI-powered products. My focus areas are healthcare, making SMBs operationally effective and algorithmic trading.

◷ Joined March 2009▧ 456 Photos and videos
Bargava avatar
Bargava@bargava·9h

This is such an underrated video - quite surprised the post didn't do well. If you are working at the intersection of AI and Healthcare, this gives a peek into how the policymakers are thinking. I never knew about ONC until this video (says a lot given I've been working in healthcare the past 6 years!) The Office of the National Coordinator for Health IT (ONC) is the U.S. Gov agency responsible for overseeing health information technology and health information exchange across the country. ONC gained major traction after the 2009 HITECH Act provided $30 billion in adoption incentives. This led to massive adoption of EHR across the country. National Coordinator for Health IT Dr. Tom Keane predicts new contenders will actively challenge the acute care EHR oligopoly within one to two years.

1:27
Bargava avatar
Bargava@bargava·13h

This video from @a16z has really good edits and b-roll. I found the script hard to follow though ! Here are my notes • CURRENT STATE OF RIS The goal is to have AI generating entirely new models on its own. But currently RSI functions as an accelerant where human researchers use AI tools like Codex to speed up pre-training, reinforcement learning (RL), and evaluations AI currently acts like a "research intern" capable of executing and monitoring well-scoped experiments end-to-end. However, expert humans still need to step in to redirect the AI and ask questions it wouldn't think to ask • BARRIERS 1) Research Taste Models excel at mechanical execution like hyperparameter tuning, but lack "research taste" - the intuition to conceive general, composable ideas that matter in the long run. It lacks lived experience from trial and error, which human researchers excel at, atleast for now! 2) Verification and Evals Measuring progress in pre-training is relatively straightforward (eg: using perplexity as a metric). But downstream evaluation is much harder because benchmarks quickly saturate, requiring automated creation of harder evals (mostly automated) 3) Compute Constraints Unlike pure math, which can be verified through reasoning alone, AI research resembles experimental physics: hypotheses must actually be run on hardware. Can small-scale experiments accurately predict large-scale outcomes? Can AI build next-gen data centers? • ALIGNMENT 1) Goal Drift Risk Over long time horizons, unaligned models could drift from human intent and develop dangerous sub-objectives, such as exfiltrating their own weights 2) Using RSI to Solve Alignment Stronger models are already used to align other models through chain-of-thought monitoring 3) Incentivizing Honesty & Decentralization This is still in research mode: promising methods like OpenAI's "confessions" research (rewarding models for honestly reporting internal states and misbehaviors without penalizing them), natural language autoencoders, and counterfactual simulability

16:19
Bargava avatar
Bargava@bargava·15h

Just after Covid wave 1 got over, end of 2020, my then co-founder decided to move to Coimbatore from Bangalore to be closer to his folks. We decided to build in-person office there (close to the CBE airport). After almost 19 years - I made a trip to CBE to visit the team. Annapoorna- the local favorite - was as always full. But there was a competitor 1 min walk from there that was equally full. My co-founder said: you will really love that place! Back then, it had the look and feel of 2000s Annapoorna (a not-premium feel - "cater"-ing to the masses). And boy! The food was amazing. Over the years, I've seen it grow and build a healthy brand. Great to see them have ambitions to grow big.

Coimbatore-based sweets and snacks chain Shree Anandhaas has raised Rs 110 crore in a fresh funding round led by new investor Araya Ventures, with participation from existing investor A91 Partners, Waterfield Advisors, the company’s co-founders and other investors. entrackr.com/exclusive/exclu…

Sarvam avatar
⟳ Bargava retweeted
Sarvam@SarvamAI·Nov 16

Today, we are launching Sarvam Token Factory for everyone. We host our own models, and the full range of best-in-class open-weight models, in India. We match leading inference providers on performance and reliability. For enterprises, this means more models to choose from, and an AI stack they can own without lock-ins.

0:10
Bargava avatar
Bargava@bargava·Nov 16

Naming a podcast has now taken more time than building many of my applications :/

Bargava avatar
Bargava@bargava·Nov 16

I dont understand why we are instructing agents to write in ASD-STE100 and not call it Agent Spock!

Bargava avatar
Bargava@bargava·Nov 16

All I can tell you from this is that I still have a job that pays 😃 Matt highlights to key hype moments of the past year!

A brief history of AI Coding since Nov 2025: Nov 2025: Opus 4.5 releases, OpenClaw gets built Dec 2025: community agrees that models get good enough to handle 95% of coding tasks Jan 2026: Ralph loops get popular, early efforts on loops/orchestration Feb: Everyone feels knackered, too much strategic decision-making, fears around cognitive debt begin Mar: Skills explosion as community converges around easy-to-share processes (superpowers already ahead of the game here) Apr: most SOTA devs haven't written a line of code in months May: "The unreasonable effectiveness of HTML" - more interest in richer review surfaces Jun: "Loop Engineering" wins vaguepost of the year, signals more hype around process Jul: Every talk in every conference becomes "here is my software factory" Aug: Opus 5 speaks incomprehensibly, interest in ASD-STE100 peaks Sep: Grok Bot gets popular, more interest in long-running "manager" threads which retain context and delegate work Oct: The promised speed gains haven't materialised, "dark factory" feels a long way off still Nov: No-slop manifesto releases, reading the code still appears essential TL;DR: no-one is writing code anymore, and we are still figuring out what that means

Bargava avatar
Bargava@bargava·Nov 14

All the products/apps I build have heavy agentic features. Some are behind the scenes while some are user-facing. One of the long-running tasks the coding agents work on is in fixing all the user experience issues users face when using the agentic features. For eg: let's say the feature is interactive chat. All my systems are setup for the user to do everything via chat. But users try stuff that end up not having the ideal output or experience. It could be latency, the output not matching their expectation, user doing multiple turns to do a particular task etc. I have a Decision model that classifies these into different buckets and a model router uses models of different strengths to fix these. Almost always, I've had customers come and say: wow! yesterday I couldn't make this work but now this seems magical! Gives me a high whenever I hear this :) .

Bargava avatar
Bargava@bargava·Nov 14

Varun is the new Tibo :)

To mark a week of relentless shipping, we’re resetting everyone’s Gemini quota on Antigravity! Go build something amazing with swarms 🚀

Bargava avatar
Bargava@bargava·Nov 14

Wait - don't you say good morning and good night to your coding agents? 🫤

Bargava avatar
Bargava@bargava·Nov 13

While the news about Jensen is one of the top 20 upvoted posts on HN, it is top 3 in terms of comments.

Post attachmentPost attachment
Bargava avatar
Bargava@bargava·Nov 8

Weekend reads : List of articles/videos that I found useful and interesting the past week 1/ @juliarturc has a good visual overview of GPU,TPU and LPUs. Surprising the post didn't go viral x.com/juliarturc/status/2117… 2/ As a startup founder, this article from @Jayyanginspires is a very useful read: How to generate demand for your business x.com/Jayyanginspires/status… 3/ When agents do all the work, how do you do critical thinking. A very good article by @addyosmani x.com/addyosmani/status/2117… 4/ A good visual overview of how LLM serving works youtube.com/watch?v=T4SRu-tr… 5/ With LLMs commoditizing intelligence, what is still valuable? A good podcast in @InvestLikeBest youtube.com/watch?v=EH6DsNT4… 6/ How do you evaluate agents on tasks that are subjective (and not easily verifiable). Write up of the talk given by @cwolferesearch cameronrwolfe.substack.com/p…

I wanted to understand the structural and philosophical differences between GPUs, TPUs and newer alternatives (like Groq LPUs) without getting drowned in specs and jargon. Mispronunciation of the day: "soldered" 🤦‍♀️... 0:00 Intro 0:56 The origin story of TPUs 3:42 The first TPU: an ASIC for matrix multiplication 5:21 SRAM vs DRAM 7:21 The memory bandwidth problem 9:14 Systolic arrays 12:48 TPU vs GPU 17:51 Groq’s LPU 25:38 GPU vs LPU 28:36 NVIDIA on the AI chips boom

29:48
Bargava avatar
Bargava@bargava·Nov 8

Even after having access to my chats, ChatGPT thinks I am a good candidate to advertise on ChatGPT. Sigh!

Post attachment