
Angel investor, mostly India. Start-ups, deeptech, entrepreneurship. Instagram @jamijamwal
Mumbai
Joined September 2009
- Tweets1.1K
- Following304
- Followers2.1K
- Likes5K
krs jamwal retweeted
India's poverty reduction is the second biggest in world history, just behind China's. India is now growing at around 6.5%, which is about the same as China in 2014. Manufacturing is growing right along side services.
More people should pay attention.
noahpinion.blog/p/india-is-q…
Worth reading and mulling over...
My AI predictions for the next year
1. Fable 5.5 -> Fable 6. GPT 6.1 Astra -> GPT 7
Fable 5.5 and GPT-6.1 Astra will be the last major releases of 2026. I suspect neither will actually ship under those names
Anthropic might even jump straight to Fable 6, just as it did with the unexpected version jump when Fable 5 was released
As for Astra, I doubt it'll be the biggest model in the GPT lineup. OpenAI clearly has something more powerful up its sleeve. My guess is GPT-7 Bel
The end of 2026 will mark the beginning of a transition to an entirely new generation of models
But this is pure speculation
2. 2027 will be the year of radical architectural experimentation
Transformers, at least in their classical form, will gradually give way to more efficient architectural components.
Several major AI labs have already made significant progress in exploring alternatives. The most promising directions include Mamba, Test-Time Training (TTT), and diffusion-based approaches to token generation.
I expect OpenAI and Anthropic to introduce entirely new experimental model families based on this research. They might not even carry the GPT or Claude branding
3. We're going to witness the return of recurrent neural networks (RNNs), which were largely abandoned for years
Promising architectures like RWKV have already demonstrated that modern recurrent approaches have serious potential. In theory, they could also improve training efficiency and quality
4. There will be no RSI miracle.
Recursive self-improvement is a powerful way to automate routine research and engineering tasks, but it still depends on fundamental breakthroughs discovered and implemented by human researchers
And the worse your initial training and RL pipelines are, the greater the risk of regression from recursive synthetic training
It's a similar problem to what we see with distillation
5. New architectures will dramatically reduce the memory requirements of LLMs
Context windows will expand severalfold, while token generation speeds will increase significantly
In 2027, I expect models with context windows of 10 million tokens or more to become increasingly common
6. Haiku- and Luna-tier models will be discontinued
7. We'll see the emergence of highly efficient models with effectively unlimited usage.
It's entirely possible that Anthropic and OpenAI will release models with Opus 5.5-level capabilities that are no longer constrained by weekly usage limits
8. New architectures will revolutionize ultra-small local models capable of running directly on pocket-sized devices.
Major AI labs will develop incredibly efficient on-device models, but instead of releasing them openly, they'll distribute them under restrictive proprietary licenses to a small number of strategic partners, such as Apple, Sony, and others
9. Large open-weight models will fall significantly further behind the closed-source frontier.
Running the most capable open-weight models will still require hardware costing tens of thousands of dollars.
10. DeepSeek will pull even further ahead of its Chinese competitors in 2027
Meanwhile, China's broader AI industry will experience a major downturn. Many labs may shut down or face significant investment cuts as their economic inefficiencies become increasingly apparent
11. Many people predict that the memory and semiconductor shortage will become even worse next year
I'm more optimistic
By mid-2027, I expect a substantial decline in hardware prices and a significant improvement in component availability
12. Google will not win the AI race
Its stock will continue to fall, potentially much further
13. Apple stock will collapse
The company will fail to meaningfully adapt its devices and services to the AI era
Meanwhile, entirely new devices will emerge from smaller, lesser-known companies and AI labs, gradually erasing the boundaries between smartphones, laptops, and servers
Everything will be controlled through a single device, with a substantial portion of the computation happening in the cloud
14. Standalone apps and websites will gradually begin to disappear
Instead of navigating predefined interfaces, users will interact with software generated dynamically in real time for their specific needs
Code itself will cease to be the primary source of value
We're already seeing the first signs of this with ChatGPT and Claude generating interactive widgets directly in their responses. This trend will accelerate dramatically
15. We're about to witness a renaissance in game development
Years of creative stagnation, repetitive formulas, and absurdly bloated development budgets will finally come to an end
Within the next two years, game development costs and production timelines will be cut in half
More and more solo developers will start releasing games with AA+ production quality
Many major publishers will enter a severe crisis. Their stock prices could collapse severalfold
16. 2027 will be the most devastating year for the stock market in human history
We'll witness the decline of an enormous number of corporations that have seemed untouchable for decades
17. Early human trials of a new generation of cancer treatments will demonstrate cure rates approaching 90%
By 2029, these groundbreaking treatments will become widely available
2
54
Superb perspectives- this is a series I look forward to reading.
The Ultimate Energy Venture
What does the word "ultimate" even mean? It means the final, highest, most extreme or the most fundamental thing in a series, a process, or a category
OK, so how do we measure energy?
Everyone has their own theory about energy, and I'm not here to persuade anyone to change their mind. But even so, everyone agrees that a couple of things about energy are self evidently true.
1. Energy is good, more energy is better. Famously there are no industrialized nations with low/weak energy consumption.
2. See above.
Therefore the "Ultimate Energy Venture" is just the biggest play, it's scale baby. More is more.
What is the absolute, the largest thing that is possible to build on the planet Earth? What does the towering zenith of scale look like? What are the maximum physical limits of machine size that our 6 zettatonne wet rocky planet allows us to build?
Well I'm glad you asked. Historically...
The heaviest self-powered land vehicle is NASA's Crawler-Transporter Two, it's 3,020 tonnes.
The heaviest self powered sea vehicle is the Pioneering Spirit it's 403,300 tonnes. The crawler is rounding error.
If you build stuff that floats, you can build stuff that's 100X bigger. This is because you can lift and float things at sea that are impossible to lift or handle on land. Building on land is nerfed, always has been. Nothing there floats. Once you have built something big at sea, you never go back.
The other important factor for building a massive energy machine is that you need a heat sink that is capable of allowing you to close the maximum scaled thermodynamics. Well guess what?
The ocean is a 1,104 times bigger heat sink than the atmosphere that exists around your land based normiewidget. That means the amount of energy required to raise the atmosphere by 1.5C would raise the ocean temp by 0.00C.
Further... air is an insulator!
What are we even doing here?
Only a moron would build the ultimate energy play on a hard deck (land) surrounded by air. From the correct perspective, which is planet level perspective, you would obviously build the maximum scale energy machine to float in your indefatigable heat sink.
Water is 3,500 more effective for heat transfer than air, so you want your ultimate heat sink to be water. The square-cube law amplifies this massively the reason a blue whale is 200 tons whilst an elephant is 5 tons is because a 200 ton elephant would overheat and die, it's impossible for 200 ton mammal to dissipate warm blood heat in air.
The ultimate heat sink interface gives you previously impossible scale and fundamentally changes both the vertical scalability and the economies of scale.
We have established that at sea we can build 100x bigger by mass, and have 1,000x the thermal emission v's what we can do on land.
Is the sea scarce? pfft
The planet is 71% open ocean.
Human civilization builds all of our stuff on the most expensive 10% of the 29% of the planet that is suboptimal for thermodynamic engines. Humans have got it all wrong. We're not using the planet properly. We're not building the optimal things. We are guilty of lazy thinking, building what is familiar because we are bipedal apes, and not what is the engineering optimal for this particular oceanic rocky planet.
From the optimal perspective, you see the opportunity optimally. Apparently this is rare.
Now seeing is one thing, doing is quite another. But yet again, some people have already done this. We've done it so many times we're good at it. We're so good at it, we do it well. Some of us know shipbuilding, and offshore operations, and subsea operations, and yes the special 'hot rocks' operations too.
Some of us have already spent billions of dollars in this supply chain. Some of us already have all the connections, the competence, the know how. You can doubt me, but there are already several multi-trillion dollar sovereign wealth funds whose very existence proves we know what we are doing.
We know how to launch it, we know how to execute it.
So that's what we're doing.
(and I'm going to tell you all about it here on X, and you're all going to benefit)
1
42
krs jamwal retweeted
Good to see Indian investors coming to for builders.
Its indeed a great time to BUILD in India, for the world.
Keep it up sir @jamijamwal
A commonly found gut microbe manufactures a molecule that is capable of slashing cholesterol levels and shrinking plaque in arteries
go.nature.com/4y50J5X
krs jamwal retweeted
Looking at OpenAI releasing hundreds of mathematical results, I think we really have to lead with empathy here.
When you’ve spent years learning a craft, whether it’s writing code, creating illustrations or doing mathematics, and suddenly a lot of what you do seems to be a prompt away, of course that’s disorienting. Your craft is how you define yourself, what you’re proud of, where you find flow, sometimes the thing that gives your day meaning. It’s a huge part of your identity.
And I get how this can lead to a pretty nihilistic feeling, like you wake up one morning and think, okay, if everything can just be prompted into existence, what am I supposed to do now? What am I getting better at? What is all this effort for?
Software engineers have been grappling with this for a while, and I expect more people will experience it as AI keeps accelerating. Writing a book, creating a movie, maybe someday helping solve a disease, more and more things we’ve spent our lives learning could feel a prompt away.
I think that progress is amazing, and I’m 100% confident we’ll find harder problems to work on and new crafts to fall in love with. But knowing that intellectually doesn’t make the sense of loss disappear, especially when the change happens this fast and feels this brutal.
We can celebrate the breakthroughs and still care about the millions of people trying to figure out where they fit and what gives their life meaning in a world becoming this intelligent. “You can do more now” might be true, but people also need time to process what they feel they’ve lost.
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
krs jamwal retweeted
I am surprised that people are surprised this is how I prompt Claude.
Talk to Claude the way you would a coworker. There's no secret to prompting. There's no need to be overly scaffolded or prescriptive for most tasks -- give Claude a goal, and it will figure it out.
Back in the Sonnet 3.5 days, your prompt mattered a lot. Nowadays, it's much more important to communicate to the model:
1. What you want it to do
2. How much effort you want it to spend
3. How it should verify that it did the right thing
krs jamwal retweeted
Also this terrific chart. All from: epoch.ai/publications/the-pl…
krs jamwal retweeted
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.
Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
krs jamwal retweeted
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
krs jamwal retweeted
Half of America's WW2 wartime research budget went to three institutions. MIT, Harvard, Caltech.
Guess who were the men handing out the money through the govt grants? The SAME people who ran MIT, Harvard, Caltech etc.
Their chairman, Vannevar Bush, had been dean of engineering at MIT. He ran one of America's biggest private research funders. He had even co-founded Raytheon on the side.
This is the nature of the conflict of interest being cooked up about the RDI scheme.
- America has been having this argument since 1945.
-IT HAS NEVER SOLVED IT BY REMOVING THE INSIDERS
-When they did remove insiders the whole thing BACKFIRED.
Let's relook at Vannevar Bush's story.
-Bush's OSRD gave America radar, penicillin at industrial scale, the proximity fuse etc.
-64 per cent of the money went to ten institutions.
-Close to half went to three institutions
-Those three were EXACTLY where the decision makers came from
What America decided:
-Keep the insiders.
-Write their interests down. India already does something like this.
MORE RECENT EXAMPLE - Q-Tel:
-CIA backed VC fund.
-nearly half its trustees had a financial link to a company it funded.
-The portfolio includes Palantir and Anduril
Where India actually is:
-India spends 0.84 per cent of GDP on R&D.
-The US spends 3.4 per cent.
-China 2.6 per cent
America kept the insiders, it made them declare in writing before the vote but it judged the portfolio after a decade.
Good explainer on how to think about the RDI related non-controversy, and an opportunity to learn from history. Do read Raghavan's story today in @SwarajyaMag 👇
swarajyamag.com/politics/rdi…
krs jamwal retweeted
You may have been told to watch this video about the OpenAI AI hack. You really should, even if you don't usually care about tech stuff.
If nothing else, click this link to the 18 minutes in & see how the agents spoke with each other. Its eye opening. youtu.be/87DyyMV0kCY?si=DMfq…
krs jamwal retweeted
At #cambridge wt @Jeffrey_Archer #MaryArcher celebrating ther Diamond #WeddingAnniversary @ historic #OldVicarage wt #BritishPrimeMinister #JohnMajor #IanBotham #Hollywood’s #KathleenKennedy #FrankMarshall of ET Indiana Jones #JurassicPark #Bourne Identity & husband @jamijamwal
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
openai.com/index/hugging-fac…
krs jamwal retweeted
𝗘𝘃𝗲𝗿𝘆 𝗲𝗻𝗱𝘂𝗿𝗶𝗻𝗴 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗶𝘀 𝗯𝘂𝗶𝗹𝘁 𝗼𝗻 𝗱𝗲𝗰𝗶𝘀𝗶𝗼𝗻𝘀 𝘁𝗵𝗮𝘁 𝗼𝘂𝘁𝗹𝗮𝘀𝘁 𝘁𝗵𝗲 𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝗰𝘆𝗰𝗹𝗲.
Today, at our Annual Client Partner Conference 2026, TCF Connect, the conversation returns to this theme through the lens of India’s technology infrastructure moment.
The conference will feature a keynote address by @kris_sg , Co-Founder, @Infosys, followed by a panel discussion on “From Ambition to Architecture: India’s Technology Infrastructure Moment” with @neelkanthmishra, Chief Economist,@AxisBank; @jamijamwal, Former ED, Tata Industries Limited; @sanjaynekkanti , Founder and CEO,@DhruvaSpace; and Sharad Sanghi, Co-Founder and CEO, Neysa Networks.
#TCFConnect #AnnualClientPartnerConference #BuildingEnduringBusinesses #IndiaInvesting
krs jamwal retweeted
SGLT2 inhibitors are diabetes drugs that work by making your kidneys dump excess glucose into urine. A new meta-analysis covering 284 patients just found they do something unexpected. They shrink the layer of fat wrapped around your heart.
This isn't the fat you can pinch. It's called epicardial adipose tissue, and it sits directly on the surface of your heart, sandwiched between the heart muscle and the protective sac around it. In some people, this fat layer makes up 20% of total heart weight.
For years, researchers thought this fat was just cushioning. It's not. This tissue actively secretes inflammatory molecules that seep directly into heart muscle, contributing to coronary artery disease, irregular heart rhythms like atrial fibrillation, and heart failure progression.
Studies show that people with more of this heart fat have higher rates of heart attacks, strokes, and heart failure, independent of how much they weigh overall. It's not just a sign you're carrying extra weight. It's a distinct problem with direct effects on your heart.
Before now, the only interventions proven to shrink this fat depot were extreme: very-low-calorie diets, intense exercise programs, or bariatric surgery. Standard weight loss approaches showed minimal impact on this specific fat.
SGLT2 inhibitors became blockbuster diabetes drugs because large trials showed they prevent heart failure hospitalizations and reduce cardiovascular death. But nobody was sure exactly how a kidney drug was protecting the heart. Researchers suspected it might be doing more than lowering blood sugar.
This meta-analysis searched major medical databases for studies that measured the fat around the heart before and after SGLT2 inhibitor treatment using CT scans, echocardiograms, or cardiac MRI. Eleven studies qualified, including 284 patients measured with precise imaging.
When comparing SGLT2 inhibitors against standard diabetes medications, the heart fat reduction was dramatic. An effect size of -1.64. In statistical terms, that's a large difference. People taking SGLT2 inhibitors saw substantially more heart fat reduction than those on conventional treatments.
Even looking at before-and-after measurements in the same people, SGLT2 inhibitors consistently reduced heart fat with an effect size of -0.62. This held true whether measured by ultrasound thickness or volumetric 3D imaging. The effect was real across different measurement methods.
Here's what stood out: the heart fat reduction was much larger than what you'd expect from the modest weight loss these drugs cause. SGLT2 inhibitors did cause small drops in body weight and blood pressure, but the shrinkage of heart fat was disproportionately large.
This suggests the drug is specifically targeting this fat depot, not just causing overall weight loss. Some fat deposits around organs respond differently to metabolic interventions than subcutaneous fat under the skin. This appears to be one of those cases.
The mechanism likely goes beyond just urinating out glucose and losing a few pounds. These drugs appear to reduce the inflammatory signaling coming from the fat tissue around the heart, which in turn reduces scarring in heart muscle and prevents harmful structural changes.
Several studies in this analysis measured inflammatory markers in the blood. Molecules like TNF-alpha that promote inflammation. They found these markers dropped alongside the heart fat reduction. The fat wasn't just shrinking, it was becoming less metabolically active and less inflammatory.
Your heart is constantly influenced by the tissue surrounding it. When that tissue is inflamed and secreting damaging molecules, it creates a hostile local environment. When that tissue shrinks and quiets down, the heart operates under less metabolic and mechanical stress.
The cardiovascular benefits of SGLT2 inhibitors have been clear in clinical trials for years. Fewer heart failure hospitalizations, lower death rates. This meta-analysis helps explain one mechanism: they're targeting a fat depot that sits millimeters from heart tissue and directly contributes to heart disease.
Until now, meaningful reduction of heart fat required extreme interventions that most people can't sustain. SGLT2 inhibitors appear to achieve targeted reductions through a mechanism that doesn't require drastic calorie restriction or surgery.
Doctors are already choosing diabetes medications based partly on heart protection, not just blood sugar control. These findings add another layer. These drugs measurably shrink the inflammatory fat surrounding the heart that contributes to disease progression.
A drug designed for kidney glucose control protects the heart by shrinking metabolically toxic fat tissue wrapped around it. The question is whether starting these drugs earlier, before this fat accumulates and causes structural damage, could prevent heart disease before it starts.
krs jamwal retweeted
AI:AC hypothesis is a solid hypothesis, as it’s probably the best proxy for a country’s ability to build out new energy infrastructure.
USA built 80GW of new energy generation for the AC build out in 70s and 80s.
Globally there are 4 technologies that will keep driving electricity demand higher over the next 15-25 years:
AI - Artificial Intelligence
EV - Electric Vehicles
AC - Air Conditioning
GA - General Automation (the actuators)
These four will play out at different times, in different orders, in different regions.
AC and EV are sigmoids under demographic ceilings but AI and GA are sigmoids that really escape any demographic limits.
Currently global GDP is about $125 trillion and the world spends about $3.5 trillion annually on electricity. Of the $3.5T about $1 trillion of that is transmission and about $2.5 trillion is generation.
The consensus is that the world needs to double electricity production from 30,000TWh/yr to 60,000TWh/yr over the next 20 years.
Now doubling the volume doesn’t mean electricity spending goes from $3.5 trillion to $7.0 trillion because this is demand driven expansion it’s going to happen under supply constrained pricing.
This is what we are already seeing with Chips and Memory suddenly jumping to 70% margins.
In the medium term, electricity transmission + generation is going to jump to 70% margin in free market conditions.
Demand is expected to double and pricing will triple (shift from 5% margin to 70% margin).
This means the global economy will WANT to spend $20 trillion on electrons.
This is what the math behind an industrial revolution looks like, if you have studied the economics of past industrial revolutions you can learn how these things play out. You can’t get perfect foresight, but certainly directional and scalar foresight.
There will be energy companies that do trillions of dollars of revenue at large margins. I expect that 30-40 years from now, there will be 12-20 companies that represent 60-70% of world GDP.
I think the most interesting thing is to figure out what those 12 companies might be, and then to go out and try to build one of them.
I think (one persons opinion) under the old regime electrons were about 3% of global GDP, but under the new technological regime, post industrial revolution, electrons will be about 10% of global GDP.
Electrons will be the general substrate of the economy, and by 2040 economic size will shift from being demographic limited to electron limited. What makes a big country a big country is about to shift from population to energy.
krs jamwal retweeted
If you want to know a lot about a person's brain health, it's embedded in a retina photo, deciphered by AI
thelancet.com/journals/landi… @LancetDigitalH



































