It is. Is it?

I heard Yann LeCunn recently say, and I am paraphrasing, that for AI to be truly meaningful, it must have the ability to predict outcomes of its actions and adjust itself accordingly. Use of AI for reasoning - work with us to brainstorm, build on ideas - is one of the most meaningful uses of AI. However, often times, AI ends up becoming an echo chamber without us realizing it. It can make your noise sound like music. You may end up loving your voice, singing louder and generating more noise, until the audience tells you they can't bear it.

Working with AI to test and build ideas can often take you too far down wrong paths. Most of these LLMs, if not all, are very good at offering reinforcement to your thought process, adding layers of rationalization, generating confident and seemingly valid grounds for various elements of solution options that you piece together, and then shaping for you the artifacts that you feel proud for having created - and your LLM doesn’t mince words in making you feel so. And yet, if you started with some wrong premise or assumption that was central to the whole pursuit, and built everything on top of it, the process would only lead you farther away from any meaningful solution to the problem you set out to solve.

Posted on LinkedIn on 02 June 2026.

AI as a 'resource' for Research

Impact of AI on research is the most talked about topic in most academic circles and research conferences these days. In one such discussion during the India Strategy Conference (ISC2025) at Indian School of Business this year, I raised a concern that if most research gets 'powered' by AI - which is increasingly the case - the quality of research outcomes will be increasingly dependent on the kind and version of AI one has employed. As AI becomes a resource, ability to carry out high quality research becomes a question of access and affordability.

I was countered with an argument that AI is an equalizer. In one sense it is, as it allows one to try his/her hands at high quality research without the depth of skillset that research demanded earlier. But beyond the basic level, inequality starts taking shape because the resource is limited and controlled. Claude Haiku 4.5/Sonnet 4.6/Opus 4.7? ChatGPT Free/Go/Plus? Gemini Free/Plus/Pro/Ultra? We all know that the free ones are extremely limited in capability and reliability. Paid ones have variants offering different levels of capability depending on how much you pay. The pricing is equal worldwide, but the purchasing power isn't. So users are forced to settle, and work with whatever is the best they can get their hands on.

Research outcomes have always been constrained by access to resources. But now, with 'intelligence' as a resource, the race can get too exhausting for many - some running on sand, some even on water - while a few cruising on firm surfaces, breezing through in skates that don't even make a sound. The bodies may be equal, the race still isn't.

Posted on LinkedIn on 27 May 2026.

Claude and its usage limits

I appreciate the makers of Claude for being so thoughtful and responsible to have built ‘work-life balance’ into it. And neatly optimized it for the $20 subscription - it allows the right amount of usage that exhausts - not too soon, not too late - then making you wait to watch reels, until you can resume using intelligence - the artificial one.

Originally posted on LinkedIn on 25 May 2026.

Short-termism is Sticky - Moving from Principal-Agent to Principal-Principal Conflicts

Now this is the other side of the corporate short-termism debate that has gained little importance. If QoQ disclosures to shareholders is forcing companies to think short-term, moving to a less frequent disclosure can create information asymmetries that can harm retail investors / minority shareholders who rely only on public disclosures and media with wide reach. The measure intended to alleviate short-termism driven by principal-agent dynamics is probably going to amplify short-termism driven by principal-principal conflicts!

I recommend reading the full letter below from wallstreetbets community to SEC!

Originally posted on LinkedIn on 16 May 2026

I wonder...

I am sipping coffee at Starbucks, trying to work, and ended up 'wondering' - May be it's not AI that's making us useless... May be it's LinkedIn. It has conditioned a whole bunch of people to start 'wondering' after every little thing they do, or don't, or just go through, or don't... An hour of wondering per minute of activity - that might as well be wondering - then an hour trying to post on LinkedIn what we wondered, often by getting AI to write about it beautifully... Then a few hours checking likes and comments, wondering about those. We even wonder about wondering. Many are wondering - even worrying - whether they are wondering enough. People have started wondering together. There are even courses to help you do high quality wondering. A cursory scan of LinkedIn will tell you that we are certainly getting better at it. AI is playing a huge role in wondering lifecycle management (WLM). It's even executing some of the steps in the workflow. Can it do the whole of it? I wonder. Soon, perhaps. But we should make sure we are always wondering 10 times better than AI. A lot of us post our pics to go along with our wondering narrative. AI can't. But then it can generate a prettier human picture than any of our wondering faces - I wonder whether that would matter. I wonder whether hallucinating by AI qualifies as wondering of some sort. Or is it just a couple of steps behind wondering? Should we worry about it, I wonder. We have to outwonder AI, as wondering is all we got left. I wonder whether we can do it. We have to. It's our 'wonderful' world afterall! But then, I wonder...

Originally posted on LinkedIn on 12 May 2026

Is AI taking away the beauty from research?

Is AI taking away the beauty from research? My response on a group chat of fellow research scholars when this topic came up - “… we must understand what research truly is - a pursuit of reality, in bits and pieces at a time… tools se processing ya synthesis ko research maan lo to hamara stata ya R use karna also takes away the beauty of doing it on reams of paper… who among us has any clue of the math that went in to create some of the models we so liberally apply without the least bit of understanding… we’ve already gone past levels of beauty many beholders would be ashamed to see humans bypass”. The fear, I guess, is the complete redundancy of the myriad capabilities we develop as we learn to do ‘research’ within conventional and widely accepted frameworks, formats and methodologies. This pushes us to think beyond even with regard to those paradigms. Genuine research is constant search for novelty. We should be excited to be equipped with a capability that allows us to venture father and deeper, and unearth the true nature of reality.

Having said that, we can’t deny that research as a profession is going to get deeply and adversely affected, as most researchers will struggle to adapt and evolve with the emerging transformation.

While the top 1% are seeing new paths for themselves, to make it rewarding for the 99% to pursue research, we need shift of perspective and also shift of foundational capabilities. How should that be approached?

Originally posted on LinkedIn on 02 May 2026.

Too much too soon? Lessons from the Wind Industry

Wind turbines became popular and underwent rapid evolution during the 1990s and early 2000s, fueled by climate change concerns that motivated exploration into alternate, clean sources of energy. Governments recognized it and incentivized investments. The technology had existed before, but had potential for massive upgrade to achieve much higher efficiencies. Engineering and R&D in this area got a boost. Every couple of years companies started launching new turbines with higher MW ratings.

For a wind turbine to generate maximum energy, it has to be placed on a site with good wind conditions. In that, all places are not equal. For example, in India, there are places where wind profiles are strong, and others where they are weak. Also, wind turbines need to be in relatively uninhabited places because they generate sounds at frequencies that are not healthy. They need to be spaced apart adequately so that they don't interfere with each other. They are massive, and are installed in large numbers to develop what are called 'wind farms'. They must stay put and be operational for 20+ years for the investments to make sense.

Now you want to place your best machines in the best places. But the best machine is best only at a point in time. While the best place is always the best. And this is where the most logical approach is also a bit silly. The best places are taken up pretty early with machines that were best at that time. However, in an emerging technology-intensive industry that is getting a lot of attention, there's constant race to develop better machines. And each upgrade makes the earlier one seem like a toy. But the irony is that the toy got the better playground. Since the best wind sites are taken, the new, better machines have to settle for poor sites.

The industry tried to mitigate this in two ways. (1) Move offshore, to find great wind conditions - but offshore has its own challenges, (2) Develop turbines for poor wind conditions - which is like developing cars for poor roads as the highways are inaccessible.

As the industry matures, the turbines in great sites get old and up for replacement. And at that point, you can put your new ones in the best places again.

It's a wave, and there's little you can do to beat it, especially in a fiercely competitive market.

Now draw parallels and think about investments in AI based solutions. In a scramble to get onboard are companies investing too much too soon? Are companies creating legacy issues for themselves by jumping to adopt while the technology and its capabilities are evolving too fast? Is it FOMO or smart investing for efficiency and value? May be the truth lies in between.

Most AI startup ideas I came across a couple of years ago based on the shape of LLM's back then are obsolete now. It's probably how any new tech waves emerge and then gradually stabilize - short and quick early on, long and slow as they wane.

Originally posted on LinkedIn on 30 April 2026