Nikhil Kumar

Presenting, @godrift_ai: Claude code for robotics simulations. I’ve been building robots for 10+ years and most of my time was spent fighting the tools and frameworks around it, not the actual robot use case. I didn't get into robotics to feel like an engineer who is stuck debugging dependencies. I got into it because physics and the human brain fascinated me, to build things that move through the real world. At some point I stopped accepting that the setup was just "part of the job." It isn't. It's just friction nobody fixed. Yet.

Once a great man said: "I believe our generation was born too late to explore the Earth, and too early to explore the stars, but born just in time to solve robotics." - @DrJimFan

Badges like college, degrees or social media presence can help to provide a cold start, but that's not even 0 to 1 for a startup. Badges that actually matter - Patience, Courage, Curiosity, Hard work, Risk taking, Storytelling.

@Dan_Jeffries1 Scared people don't make good career decisions. Scared societies don't make good policy. The cost of bad forecasting isn't abstract.

Jensen is one the smartest and most far seeing folks the world. "If an AI scientist warns people that AI is going to permeate across radiology and radiologists are going to get wiped out, it might seem helpful but it's hurtful. If we convince everybody not to be radiologists and we now need radiologists, that actually is hurtful to society. "It is hurtful to convince all the young college graduates not to study software engineering because we are going to need more software engineers than ever. That's hurtful." "Scaring people with nonsensical things, which are not going to happen, that this is an existential threat, there's a 20% chance that is is existential, that's ridiculous. "That it's going to wipe out 50% of college level jobs. "That is it going to completely destroy democracy. "These kinds of comments are not helpful. They are made by...CEOS. And you become a CEO, maybe you adopt a God complex and somehow you know everything." Brutal. And right.

@TheGeorgePu the problem is their jobs were built around a constraint that just disappeared.

Amazon is firing 30,000 people and hiring 11,000. I sat there and thought - why? Why fire and hire at the same time? So I dug in. The 30,000 cuts? Mostly middle managers. The 11,000 hires? Individual contributor engineers. Amazon thinks there are too many managers. AI can replace your managers. Now it all made sense. AI didn't kill the developer job. AI killed the manager job. Think about it. When one engineer with Claude does what 5 used to do. You don't need the manager who managed the 5 people. You don't need their manager either. The org chart isn't shrinking from the bottom. It's collapsing from the middle. If your job is to manage people. AI is coming for you.

@rohanpaul_ai tbh the inexpensive start framing is more relevant now than when he said it. the excuse of needing capital to validate is gone.

What Mark Cuban looked for in entrepreneurs before investinng in his early days. "I want people who are true entrepreneurs that realize that sweat equity goes a whole lot further than money, and then brains and effort go a whole lot further than connections. I look for an original idea, something that's not a "here and now," but a "future." Because if you've got the "next this" or the "next that," I'm not interested. I don't want the "next one" because it's not a lot to invest into, as there already is a MySpace and there already is a YouTube. I'm looking for the next thing. If you can tell me why and how you're going to compete with Google and Microsoft—because you know if it's a good idea, they're going to compete with you—I'm interested in it. And if you tell me that we can start up inexpensively and grow it, as opposed to wanting to raise $50 million so you can hire your whole family, I'm interested."

@r0ck3t23 The gap between what the brain can do and what the body can execute has always been the hidden tax on human output.

Elon Musk just made the human body optional. Not in theory. Not in a lab simulation. In a living, breathing person. Neuralink’s first patient is quadriplegic. Can’t move his arms. Can’t move his legs. Nothing below his neck responds. Right now he’s playing video games. Downloading software. Navigating a screen with precision. Not with his hands. Not with his voice. With his thoughts. Musk: “He literally is just controlling the screen. He can play video games, download software, really anything you can do with a mouse, just by thinking.” A man whose body failed him just bypassed it entirely. His brain said move. His body said no. So Neuralink removed the body from the conversation. This is not a medical breakthrough. This is a philosophical detonation. All of civilization was built on one assumption. The body is the instrument. Hands build. Legs carry. Mouths speak. The brain just gives the orders. Neuralink inverted the entire hierarchy. The brain is no longer commanding the body. It’s replacing it. Every limitation you’ve ever felt was your body failing to keep pace with your mind. Your fingers can’t type as fast as you think. Your mouth can’t speak as fast as you process. The body was never the tool. It was always the bottleneck. And Musk just proved it with a man who now moves through the digital world faster than most of us move through the physical one. We didn’t watch a patient get treated. We watched the first human operate without a body. The rest of us just haven’t realized we’re next.

@aakashgupta Each wave got killed not because the idea was wrong but because the hardware wasn't ready. the idea was always right.

Marc Andreessen just put AI on an 80-year clock. The math isn't a metaphor. The first neural network paper dropped in 1943, when McCulloch and Pitts modeled neurons as logic gates and proved networks could compute anything a Turing machine could. Then 79 years of trying. Frank Rosenblatt built the Mark I Perceptron in 1958. He told The New York Times it would walk, talk, see, and reproduce itself. It learned to recognize triangles. Minsky and Papert killed the field in 1969 with one book. Showed perceptrons couldn't compute XOR. Funding collapsed. First AI winter. Backprop came back in 1986. Hinton, Rumelhart, Williams. The field re-funded around expert systems and LISP machines. Both blew up. Second AI winter. Hinton, LeCun, and Bengio kept publishing through the cold. AlexNet won ImageNet in 2012 by a margin that broke the leaderboard. Transformers landed in 2017. ChatGPT crossed the threshold in November 2022. 79 years from theory to product. Marc's comp list is doing real work. Steam engine, microprocessor, electricity. Not the internet. Each one had a long incubation, then a deployment phase that took another 50 years to hit saturation. Steam: Newcomen in 1712, Watt's commercial engine in 1776, mass deployment in the 1830s. Electricity: Faraday in 1831, Edison's Pearl Street in 1882, half of US homes by 1925. Microprocessor: Intel 4004 in 1971, billion-unit market by 2000. Each hit ubiquity faster than the last. Compute, capital, and distribution kept compressing the timeline. If AI is bigger than the internet and behaves like electricity, the technology question is settled. The hard question is whether power, chips, and labor can absorb it without breaking. Three years past the moment the math started working. Seventy-seven to go.

@haider1 whoever builds the first genuinely agent-native framework owns a very interesting piece of the stack

Andrej Karpathy explains why infrastructure must be built for AI agents: "most docs, frameworks, and libraries are still written for humans, when AI agents should be the primary audience the next shift is agent-first infrastructure, where systems are built for LLMs to understand, act, and automate without humans"

@aakashgupta The crazy part is it's happened consistently enough to be a pattern.

Anthropic is now at $44B ARR, up from $9B in December. What if Anthropic completes the 10x two more years in a row? $100B this year end. $1T next year end. It’s already done it the 3 years prior. People may be failing to understand that AI really is on an exponential. We keep expecting it to end, but the signs aren’t there. What if things just continue?

@rohanpaul_ai What changes isn't just company size. it's who gets to build.

Sam Altman: "We're going to see 10 person billion-dollar companies pretty soon. In my little group chat with CEO-friends, there's this One-person billion-dollar company, which would have been unimaginable without AI, and now it'll happen."

Great people in an organisation should be valued not just by compensation but also understanding their mental health. The best leaders are successful builders but also compassionate.

Hiring fast but few is the real skill. The few can't just be great at what they do, they have to embody the values you want the whole company to run on. Because more than skills, getting everyone to strategise, sync, and execute together is the actual goal.

@aakashgupta 42,000 people holding more market value than tens of millions is the clearest signal yet that we're pricing intelligence infrastructure differently than any asset class before it.

Nvidia just passed India's entire stock market. One company in Santa Clara is now worth more than every publicly traded business in a country of 1.4 billion people. Nvidia closed at $5.05 trillion. India's listed market sits around $4.6 trillion. That includes Reliance, TCS, HDFC Bank, Infosys, ICICI, Bharti Airtel, and roughly 5,000 other companies. Combined, less than one chipmaker. Nvidia has 42,000 employees. India's listed companies employ tens of millions. The composition is what makes this remarkable. India's $4.6 trillion is built on 1.4 billion consumers, IT services exports to 50+ countries, manufacturing, banking, energy, and pharma. Thousands of customer relationships across hundreds of categories. Nvidia's $5 trillion is built on data center GPUs. Data center is 90% of revenue. Four customers account for 61% of total sales. Hyperscalers alone are over half of data center revenue. Goldman Sachs projects those buyers will spend $527 billion on AI capex in 2026. The valuation is a forward curve on a handful of capex budgets. The last comparable concentration was Cisco in March 2000. Cisco hit $555 billion as the most valuable company in the world. Telecoms and dot-coms had to buy its routers and switches to build the internet. The narrative was unbreakable. Cisco's market cap fell 86% over the next 30 months. 26 years later it still hasn't recovered to its dot-com peak. The internet got built. The fiber stayed in the ground. Cisco just wasn't the company that captured the value on the next loop. Nvidia's fundamentals are nothing like 2000-Cisco. Sales hit $215.9 billion in fiscal 2026, up 65% year over year. Data center alone grew 68% to $193.7 billion. The earnings exist. The structure of the valuation has the same shape. A small number of buyers growing a small number of capex budgets at a steep rate. Change one variable and the math changes everywhere. India's $4.6 trillion is an economy. Nvidia's $5 trillion is a contract on four customers.

@haider1 Interesting!!

demis is 100% on a 5-year AGI timeline, i.e., 2030 something has definitely changed since the start of 2026, as i've noticed he has now said '2030' five times, instead of 5-10 years his quote from yesterday: "2030... i've been pretty consistent about that" his AGI definition: "true agi should match human-level general intelligence and be strong enough that even experts would need months to find a serious weakness"

@kimmonismus A commercial AI lab now has a waitlist managed by the white house. That sentence didn't exist a year ago.

The White House blocked Anthropic from expanding Mythos access beyond ~50 organizations to ~120. Not because the model is too dangerous. Because the government wants priority. Officials literally worried more customers would hamper their own ability to use it. Frontier AI just became a rationed strategic asset.

@cryptopunk7213 Two frontier labs hitting the same benchmark independently is better validation than any internal eval. The bar is real.

oh shit its official: gpt 5.5 is as powerful as claude mythos, it pulled off a 12-hour expert-level cyber security attack in <11 minutes... for $1.73 this is the type of study that dismisses the bullshit: > they tested gpt 5.5 in a cyber-attack simulation (same one mythos was tested on)... it fucking crushed. > 73% success rate across all attacks. mythos hit the same bar but it gets better... > on a task that took a human expert 12 hours to complete, gpt 5.5 did it in 11 minutes and cost $1.73 🥹 the trend is pretty clear: the secret sauce for frontier models is compute, openai secured a LOT of it and now might leap-frog anthropic (near-term) there is no ceiling on compute trained models. more tokens in = AGI-like models.

@haider1 What makes this benchmark sharp is it tests for the rarest thing, not knowing the answer, but asking the right question nobody had asked yet.

Demis Hassabis proposed a benchmark for scientific AGI: the "Einstein test" Train a system with a knowledge cutoff at 1901, then test whether it can independently rediscover what Einstein did in 1905, including special relativity Once it can, we're on the verge of genuinely novel invention

@MarioNawfal The cluster is the moat here more than the model. whoever controls the compute at that scale controls the pace.

🇺🇸 xAI is training 7 AI models at the same time on the world's biggest AI cluster. And Grok 5 Max is coming with 10 trillion parameters. The current Grok has 0.5 trillion… That's a 20x jump. The gap between today and what's coming is bigger than most people realize. Source: @shiri_shh, @xAI, @Grok, @elonmusk

@aakashgupta The distribution of it is the open variable and that part doesn't solve itself.

Elon Musk just told you to stop saving for retirement. He's worth $817 billion. None of it sits in a savings account. None in bonds. None in treasuries. He owns equity in Tesla, SpaceX, xAI, and the data centers powering them. His thesis on the podcast: AI and robotics collapse labor costs to zero. Living costs follow. Scarcity ends. Saving for security becomes saving for a world that stops existing. The advice is right. The reasoning hides a constraint that determines who survives the transition. Compute, energy, and land don't go to zero. Training a frontier model already costs $500M and rising. Inference scales linearly with usage. Power demand at major hyperscalers is projected to double by 2030. Land near grid capacity is the new oil. In a labor-free economy, owners of the AI, the energy, and the compute capture the entire productivity gain. Everyone else captures nothing. Wages disappear. Dividends absorb everything that used to pay salaries. The savings rate becomes MORE important in his scenario. The form is what changes. A 401k of bond ladders and target-date funds prices in a world where labor still earns wages. That world is the one going away. The retirement vehicle that survives is equity. Tesla, Nvidia, the SpaceX-xAI complex, the energy build-out, the land under the data centers. You want to own the means of production when production stops needing humans. Elon is the highest-conviction expression of his own thesis. The richest man alive is fully invested in the assets that capture the abundance and zero invested in the instruments that don't. The retirement that becomes irrelevant is the bond-laddered, social-security-supplemented version your parents had. The retirement that matters is whether you owned compute when labor went to zero.

@rohanpaul_ai Leverage has always been the game. Capital, people, code. Tokens is just the newest form of it.

Andrej Karpathy: "To get the most out of the tools that have become available now, you have to remove yourself as the bottleneck. You cannot be there to prompt the next thing. You need to take yourself outside the loop. You have to arrange things such that they are completely autonomous. The more you can maximize your token throughput and not be in the loop, the better. This is the goal. So, I kind of mentioned that the name of the game now is to increase your leverage. I put in very few tokens just once in a while, and a huge amount of stuff happens on my behalf." --- From @NoPriorsPod YT channel (link in comment)