Rohan Paul's banner
Rohan Paul's profile picture

Rohan Paul

@rohanpaul_ai • 152,110 subscribers

Compiling in real-time, the race towards AGI. The Largest Show on X for AI. 🗞️ Get my daily AI analysis newsletter to your email 👉 https://t.co/6LBxO8215l

Shorts

Fable 5 absolutely crushed the HTML5 physics contest, but cost 6x more than Opus 4.8 and 39× more than GLM 5.2 in that test. Test was done on atomic[.]chat, a desktop app that runs LLMs locally. The test asked 4 models to generate self-contained canvas demos with believable motion and collisions. The scenes were not simple animations because every crash needed gravity, force, timing, and contact handling. Outputs: - Fable 5: 62,158 tokens, $3.12 - GPT 5.5: 37,753 tokens, $1.14 - Opus 4.8: 22,280 tokens, $0.56 - GLM 5.2: 36,246 tokens, $0.08

Fable 5 absolutely crushed the HTML5 physics contest, but cost 6x more than Opus 4.8 and 39× more than GLM 5.2 in that test. Test was done on atomic[.]chat, a desktop app that runs LLMs locally. The test asked 4 models to generate self-contained canvas demos with believable motion and collisions. The scenes were not simple animations because every crash needed gravity, force, timing, and contact handling. Outputs: - Fable 5: 62,158 tokens, $3.12 - GPT 5.5: 37,753 tokens, $1.14 - Opus 4.8: 22,280 tokens, $0.56 - GLM 5.2: 36,246 tokens, $0.08

204,694 views

🦿Xpeng showed a humanoid robot called IRON whose movement looked so human that the team literally cut it open on stage to prove it is a machine. IRON uses a bionic body with a flexible spine, synthetic muscles, and soft skin so joints and torso can twist smoothly like a person. The system has 82 degrees of freedom in total with 22 in each hand for fine finger control. Compute runs on 3 custom AI chips rated at 2,250 TOPS (Tera Operations Per Second), which is far above typical laptop neural accelerators, so it can handle vision and motion planning on the robot. The AI stack focuses on turning camera input directly into body movement without routing through text, which reduces lag and makes the gait look natural. Xpeng staged the cut-open demo at AI Day in Guangzhou this week, addressing rumors that a performer was inside by exposing internal actuators, wiring, and cooling. Company materials also mention a large physical-world model and a multi-brain control setup for dialogue, perception, and locomotion, hinting at a path from stage demos to service work. Production is targeted for 2026, so near-term tasks will be limited, but the hardware shows a serious step toward human-scale manipulation.

🦿Xpeng showed a humanoid robot called IRON whose movement looked so human that the team literally cut it open on stage to prove it is a machine. IRON uses a bionic body with a flexible spine, synthetic muscles, and soft skin so joints and torso can twist smoothly like a person. The system has 82 degrees of freedom in total with 22 in each hand for fine finger control. Compute runs on 3 custom AI chips rated at 2,250 TOPS (Tera Operations Per Second), which is far above typical laptop neural accelerators, so it can handle vision and motion planning on the robot. The AI stack focuses on turning camera input directly into body movement without routing through text, which reduces lag and makes the gait look natural. Xpeng staged the cut-open demo at AI Day in Guangzhou this week, addressing rumors that a performer was inside by exposing internal actuators, wiring, and cooling. Company materials also mention a large physical-world model and a multi-brain control setup for dialogue, perception, and locomotion, hinting at a path from stage demos to service work. Production is targeted for 2026, so near-term tasks will be limited, but the hardware shows a serious step toward human-scale manipulation.

3,802,270 views

And Robotic hands are also evolving faster than you think 👀

And Robotic hands are also evolving faster than you think 👀

1,001,187 views

Meta open-sourced a brain-to-text system that reaches 78% word accuracy without surgery. Brain2Qwerty v2 converts non-invasive brain recordings into text with 61% average word accuracy and 78% for its strongest participant. The system reads MEG signals from a helmet, not electrodes placed inside brain tissue. 9 volunteers typed about 22,000 sentences while researchers recorded 10 hours of neural activity each. Brain2Qwerty v1 mostly mapped brain signals to single typed characters. It tries to recover characters, words, and full sentence meaning together. The system studies those brain signals and tries to turn them into the words you wanted to type. - 61% average word accuracy across all participants - 78% word accuracy for the top participant - 50%+ of sentences decoded with no more than 1 word error Performance improves as the data pile grows Raw brain signals are messy because many mental and physical processes fire at once. Deep learning handles that mess by learning patterns directly from the original recordings. A fine-tuned LLM then uses language context to repair likely word and sentence errors. This explains why the system beats earlier non-invasive methods reporting 8% word accuracy. More than half of sentences from the strongest participant had one word error or less. Accuracy also improved as training data grew, suggesting more recordings may close more of the gap.

Meta open-sourced a brain-to-text system that reaches 78% word accuracy without surgery. Brain2Qwerty v2 converts non-invasive brain recordings into text with 61% average word accuracy and 78% for its strongest participant. The system reads MEG signals from a helmet, not electrodes placed inside brain tissue. 9 volunteers typed about 22,000 sentences while researchers recorded 10 hours of neural activity each. Brain2Qwerty v1 mostly mapped brain signals to single typed characters. It tries to recover characters, words, and full sentence meaning together. The system studies those brain signals and tries to turn them into the words you wanted to type. - 61% average word accuracy across all participants - 78% word accuracy for the top participant - 50%+ of sentences decoded with no more than 1 word error Performance improves as the data pile grows Raw brain signals are messy because many mental and physical processes fire at once. Deep learning handles that mess by learning patterns directly from the original recordings. A fine-tuned LLM then uses language context to repair likely word and sentence errors. This explains why the system beats earlier non-invasive methods reporting 8% word accuracy. More than half of sentences from the strongest participant had one word error or less. Accuracy also improved as training data grew, suggesting more recordings may close more of the gap.

17,127 views

Rumors suggest Dreamina (operated by ByteDance) is preparing a smaller new Seedance release. The buzz says Dreamina Seedance 2.0 mini could land on June 15, bringing near-Seedance 2.0 quality without the same painful price tag. For creators who love Seedance but not the bills, this could be very welcome. For a while, everyone focused on raw output quality. Now the bigger question is: how many serious attempts can you make before the process becomes too slow or expensive? Better AI video for less money is always nice. #dreamina #seedance #dreaminaseedance2mini You can try it here.

Rumors suggest Dreamina (operated by ByteDance) is preparing a smaller new Seedance release. The buzz says Dreamina Seedance 2.0 mini could land on June 15, bringing near-Seedance 2.0 quality without the same painful price tag. For creators who love Seedance but not the bills, this could be very welcome. For a while, everyone focused on raw output quality. Now the bigger question is: how many serious attempts can you make before the process becomes too slow or expensive? Better AI video for less money is always nice. #dreamina #seedance #dreaminaseedance2mini You can try it here.

50,604 views

🇨🇳 In China, there are robots that double as solar panels and use the power they generate to clean other solar panels. Snow often covers solar panels at photovoltaic power stations durng winter. This robot automatically removes the snow.

🇨🇳 In China, there are robots that double as solar panels and use the power they generate to clean other solar panels. Snow often covers solar panels at photovoltaic power stations durng winter. This robot automatically removes the snow.

215,683 views

Robot's locomotion and recovery under unexpected force in real time. The recovery phase was something 😀

Robot's locomotion and recovery under unexpected force in real time. The recovery phase was something 😀

38,498 views

Now it all makes sense, Claude Sonnet 4.5 can keep its coding focus for nonstop 30 hours. And Dario Amodei just said few days back that, "The vast majority of code that is used to support Claude and to design the next Claude is now written by Claude. It's just the vast majority of it within Anthropic. And other fast moving companies, the same is true." The shift has started in all tech companies. --- From 'Axios' YT Channel.

Now it all makes sense, Claude Sonnet 4.5 can keep its coding focus for nonstop 30 hours. And Dario Amodei just said few days back that, "The vast majority of code that is used to support Claude and to design the next Claude is now written by Claude. It's just the vast majority of it within Anthropic. And other fast moving companies, the same is true." The shift has started in all tech companies. --- From 'Axios' YT Channel.

226,901 views

Andrej Karpathy just put out this tool that looks at AI's impact on job. He also deleted the original Github repo very quickly. Basically, he pulled 342 job types from the Bureau of Labor Statistics and had an LLM score each one from 0 to 10 based on AI exposure. The average exposure score is 5.3. Move the score, move the probability it will get wiped out by AI. - Software developers 9/10, - medical transcriptionists are a 10/10. - Lawyers 8/10 - General Office clerks 9/10 Basically any screen-based jobs are in trouble. $3.7T annual wages in high-exposure jobs (7+) pre-computed as ∑(BLS employment count × BLS median annual wage) over exactly those occupations whose Gemini Flash score is ≥7.

Andrej Karpathy just put out this tool that looks at AI's impact on job. He also deleted the original Github repo very quickly. Basically, he pulled 342 job types from the Bureau of Labor Statistics and had an LLM score each one from 0 to 10 based on AI exposure. The average exposure score is 5.3. Move the score, move the probability it will get wiped out by AI. - Software developers 9/10, - medical transcriptionists are a 10/10. - Lawyers 8/10 - General Office clerks 9/10 Basically any screen-based jobs are in trouble. $3.7T annual wages in high-exposure jobs (7+) pre-computed as ∑(BLS employment count × BLS median annual wage) over exactly those occupations whose Gemini Flash score is ≥7.

91,449 views

New from Ilya Sutskever He talks about AI recursively building better AI. He argues such systems could compress decades of biomedical RAG into months, erasing many diseases and extending life, yet the same loop could outstrip any safety protocol. Prediction fails because super-capable agents remain both unpredictable and unimaginable. From 'The Open University of Israel' YT Channel (Full video link in comment)

New from Ilya Sutskever He talks about AI recursively building better AI. He argues such systems could compress decades of biomedical RAG into months, erasing many diseases and extending life, yet the same loop could outstrip any safety protocol. Prediction fails because super-capable agents remain both unpredictable and unimaginable. From 'The Open University of Israel' YT Channel (Full video link in comment)

258,413 views

Hunyuan 3D-2.1 turns any flat image into studio-quality 3D models. And you can do it on this Hugging Face space for free.

Hunyuan 3D-2.1 turns any flat image into studio-quality 3D models. And you can do it on this Hugging Face space for free.

222,392 views

OpenAI's AgentKit will be so insane, build every step of agents on one platform. These visual agent builders make the whole process of iterating and launching agents far more efficient. It sits on top of the Responses API and unifies the tools that were previously scattered across SDKs and custom orchestration. It lets developers create agent workflows visually, connect data sources securely, and measure performance automatically without coding every layer by hand. The core of AgentKit is the Agent Builder, a drag-and-drop canvas where each node represents an action, guardrail, or decision branch. Developers can link these nodes into multi-agent workflows, preview results instantly, and version each setup. It supports inline evaluation so that developers can see how changes affect output before deploying. The Connector Registry is a single admin panel that manages how data and tools connect across the OpenAI ecosystem. It centralizes integrations like Google Drive, SharePoint, Dropbox, and Microsoft Teams. Large organizations can govern access and flow of data between agents securely under one global console. ChatKit provides a ready-to-use chat interface for embedding agents inside apps or websites. It manages streaming, message threads, and model reasoning displays automatically. Developers can skin the interface to match their product without writing custom front-end code. Under the hood, all these blocks use the same execution core that runs agent reasoning through OpenAI’s APIs. Workflows in Agent Builder compile down to structured instructions for the Responses API, which handles model calls, tool use, and context passing. Connector Registry handles authentication and routing for external tools, while Evals and RFT provide feedback loops that improve agents over time. This integration means developers no longer need to handle orchestration logic, model evaluation pipelines, or safety layers separately. Everything runs natively within OpenAI’s control plane with managed security, automatic versioning, and built-in testing. In short, AgentKit standardizes the entire life cycle of an AI agent—from visual design to deployment and performance tuning—inside a single unified system.

OpenAI's AgentKit will be so insane, build every step of agents on one platform. These visual agent builders make the whole process of iterating and launching agents far more efficient. It sits on top of the Responses API and unifies the tools that were previously scattered across SDKs and custom orchestration. It lets developers create agent workflows visually, connect data sources securely, and measure performance automatically without coding every layer by hand. The core of AgentKit is the Agent Builder, a drag-and-drop canvas where each node represents an action, guardrail, or decision branch. Developers can link these nodes into multi-agent workflows, preview results instantly, and version each setup. It supports inline evaluation so that developers can see how changes affect output before deploying. The Connector Registry is a single admin panel that manages how data and tools connect across the OpenAI ecosystem. It centralizes integrations like Google Drive, SharePoint, Dropbox, and Microsoft Teams. Large organizations can govern access and flow of data between agents securely under one global console. ChatKit provides a ready-to-use chat interface for embedding agents inside apps or websites. It manages streaming, message threads, and model reasoning displays automatically. Developers can skin the interface to match their product without writing custom front-end code. Under the hood, all these blocks use the same execution core that runs agent reasoning through OpenAI’s APIs. Workflows in Agent Builder compile down to structured instructions for the Responses API, which handles model calls, tool use, and context passing. Connector Registry handles authentication and routing for external tools, while Evals and RFT provide feedback loops that improve agents over time. This integration means developers no longer need to handle orchestration logic, model evaluation pipelines, or safety layers separately. Everything runs natively within OpenAI’s control plane with managed security, automatic versioning, and built-in testing. In short, AgentKit standardizes the entire life cycle of an AI agent—from visual design to deployment and performance tuning—inside a single unified system.

178,460 views

Satya Nadella: Microsoft’s latest Wisconsin AI data center keeps yearly water consumption no higher than that of 1 local restaurant. "The cooling loop is filled once and the data centre can operate effectively with zero water consumption. Daily water usage across a year is roughly equivalent to what a single restaurant would use" The mechanism is mainly about replacing evaporative cooling with closed-loop direct-to-chip liquid cooling, so water moves like coolant inside a sealed machine rather than being boiled off into the air. Hot GB200-class AI racks produce too much heat for normal air cooling, so cold liquid is pushed through pipes into the servers and across metal cold plates touching the hottest chips. The liquid enters the rack cool, absorbs heat from the chips through cold plates, then exits the rack at a higher temperature and carries that heat through pipes to a huge cooling system outside the compute floor. Microsoft says Fairwater sends that hot water to cooling “fins” beside the datacenter, where 172 20-foot fans blow air across the fins and dump the heat into the outside air. The important detail is that the air cools the water through metal surfaces, so the water does not need to evaporate the way many older datacenters use cooling towers. The cooled liquid then returns to the servers, repeats the loop, and keeps absorbing heat from the chips. In older data centers, heat is often removed partly through cooling towers. Hot water meets moving air, some water evaporates, and that phase change carries heat away. Effective, but it consumes fresh water continuously. But Firwater is a closed loop because the same coolant keeps circulating through sealed pipes: it absorbs heat from the chips, releases that heat through radiator-like fins, then flows back to the chips again. For Wisconsin Fairwater, Microsoft says more than 90% of the facility uses closed-loop liquid cooling, while the remaining portion uses outside air and switches to water only on the hottest days. ---- From "Microsoft" YouTube channel, (link in comment)

Satya Nadella: Microsoft’s latest Wisconsin AI data center keeps yearly water consumption no higher than that of 1 local restaurant. "The cooling loop is filled once and the data centre can operate effectively with zero water consumption. Daily water usage across a year is roughly equivalent to what a single restaurant would use" The mechanism is mainly about replacing evaporative cooling with closed-loop direct-to-chip liquid cooling, so water moves like coolant inside a sealed machine rather than being boiled off into the air. Hot GB200-class AI racks produce too much heat for normal air cooling, so cold liquid is pushed through pipes into the servers and across metal cold plates touching the hottest chips. The liquid enters the rack cool, absorbs heat from the chips through cold plates, then exits the rack at a higher temperature and carries that heat through pipes to a huge cooling system outside the compute floor. Microsoft says Fairwater sends that hot water to cooling “fins” beside the datacenter, where 172 20-foot fans blow air across the fins and dump the heat into the outside air. The important detail is that the air cools the water through metal surfaces, so the water does not need to evaporate the way many older datacenters use cooling towers. The cooled liquid then returns to the servers, repeats the loop, and keeps absorbing heat from the chips. In older data centers, heat is often removed partly through cooling towers. Hot water meets moving air, some water evaporates, and that phase change carries heat away. Effective, but it consumes fresh water continuously. But Firwater is a closed loop because the same coolant keeps circulating through sealed pipes: it absorbs heat from the chips, releases that heat through radiator-like fins, then flows back to the chips again. For Wisconsin Fairwater, Microsoft says more than 90% of the facility uses closed-loop liquid cooling, while the remaining portion uses outside air and switches to water only on the hottest days. ---- From "Microsoft" YouTube channel, (link in comment)

26,957 views

HBM (High-bandwidth memory) is becoming a major bottlencek for AI. “buy more GPUs” is not the only bottleneck anymore. Because "Without the HBM memory, there is no AI Super Computer" ~ Jensen Huang

HBM (High-bandwidth memory) is becoming a major bottlencek for AI. “buy more GPUs” is not the only bottleneck anymore. Because "Without the HBM memory, there is no AI Super Computer" ~ Jensen Huang

108,523 views

🇨🇳 China's patent filing rose by 67.7x between 2000 and 2024. From 26,553 patent applications in 2000 to 1.8mn in 2024. The US sits at 503K and Japan at 421K. A very different IP battlefield than the one people were used to in the 2000s.

🇨🇳 China's patent filing rose by 67.7x between 2000 and 2024. From 26,553 patent applications in 2000 to 1.8mn in 2024. The US sits at 503K and Japan at 421K. A very different IP battlefield than the one people were used to in the 2000s.

72,063 views

Very nice resource. claude-code-best-practice trending on GitHub with 25,000+ ⭐️

Very nice resource. claude-code-best-practice trending on GitHub with 25,000+ ⭐️

52,192 views

The Unitree G1 humanoid robot trying to clear snow in a parking lot. While it’s currently struggling so much, very soon it will do these tasks better than us.

The Unitree G1 humanoid robot trying to clear snow in a parking lot. While it’s currently struggling so much, very soon it will do these tasks better than us.

74,189 views

"You don't have to speak Python, or C++ or Fortran. You just can speak human." AI is wiping out all the gates that once stood in the way to build and leverage technology. How much faster would the world progress with 100M software engineers vs 2M ?

"You don't have to speak Python, or C++ or Fortran. You just can speak human." AI is wiping out all the gates that once stood in the way to build and leverage technology. How much faster would the world progress with 100M software engineers vs 2M ?

96,824 views

ByteDance just open sourced an AI SuperAgent that can research, code, build websites, create slide decks, and generate videos. All by itself. DeerFlow 2.0 (27K+ GitHub stars ⭐️), an AI system acting like an autonomous employee with its own computer workspace to research and code. Standard chatbots only generate text and forget your preferences. DeerFlow solves this by giving the AI an isolated virtual computer environment where it safely runs programs. When given a massive task, the main program creates several smaller AI assistants to work simultaneously. It also saves your past workflows so it gets smarter about your needs. DeerFlow is model-agnostic — it works with any LLM that implements the OpenAI-compatible API. Fully supports running local models on your own computer using tools like Ollama. An example - you ask for research on the top 10 AI startups in 2026 for a presentation, the lead agent in DeerFlow breaks that big job into smaller sub-tasks. It assigns one sub-agent to look into each company, another to find funding details, and a third to handle competitor analysis. These agents do all their work in parallel. Everything eventually converges, and a final agent pulls the results into a slide deck complete with custom visuals.

ByteDance just open sourced an AI SuperAgent that can research, code, build websites, create slide decks, and generate videos. All by itself. DeerFlow 2.0 (27K+ GitHub stars ⭐️), an AI system acting like an autonomous employee with its own computer workspace to research and code. Standard chatbots only generate text and forget your preferences. DeerFlow solves this by giving the AI an isolated virtual computer environment where it safely runs programs. When given a massive task, the main program creates several smaller AI assistants to work simultaneously. It also saves your past workflows so it gets smarter about your needs. DeerFlow is model-agnostic — it works with any LLM that implements the OpenAI-compatible API. Fully supports running local models on your own computer using tools like Ollama. An example - you ask for research on the top 10 AI startups in 2026 for a presentation, the lead agent in DeerFlow breaks that big job into smaller sub-tasks. It assigns one sub-agent to look into each company, another to find funding details, and a third to handle competitor analysis. These agents do all their work in parallel. Everything eventually converges, and a final agent pulls the results into a slide deck complete with custom visuals.

50,097 views

In China, 158km highway with zero humans, only AI and Robots. The Beijing–Hong Kong–Macao Expressway upgrade, where an unmanned paving and rolling fleet followed satellite-guided plans while engineers monitored remotely.

In China, 158km highway with zero humans, only AI and Robots. The Beijing–Hong Kong–Macao Expressway upgrade, where an unmanned paving and rolling fleet followed satellite-guided plans while engineers monitored remotely.

73,057 views

Videos

LIVE
1.2k

Anya Rossi

sweetdream.ai

SweetDream.aiSponsored•Livecam
Streaming Now

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

HD live stream
Exclusive private shows
1.2k viewers online

Free preview available • Premium content

Robots in China are doing it all now, even dancing on stage like pros. Here Unitree robots doing Webster flips and are performing at Chinese-American singer Wang Leehom’s concert in Chengdu.
0:39
rohanpaul_ai's profile picture

Robots in China are doing it all now, even dancing on stage like pros. Here Unitree robots doing Webster flips and are performing at Chinese-American singer Wang Leehom’s concert in Chengdu.

Rohan Paul

53,578,867 views • 6 months ago

CNBC interviewer asked Palantir CEO Alex Karp how he would defend Wall Street’s concern that AI could replicate what Palantir is doing. Karp defended by basically saying that AI companies may have great engineers, but they do not deeply understand the messy, high-stakes enterprise problems Palantir solves on the ground. ------ Alex Karp: "No one in enterprise factually is worried. I've spent all my life, for better or worse, dealing with the most complicated, most interesting enterprises. I'm on the ground floor of that, probably like no one else. Those kinds of engineers are great engineers, and I'm telling you, they don't talk to the enterprises or understand the technical challenge. If you want to manufacture a car and you need a part, or you want to send a rocket to the moon, or you want to put a missile on your adversary's head and bring America home safely, that stuff doesn't ship. And by the way, there is not a single high-end enterprise like that that would ever put that in place. That is before you even get to the cultural impasse." ---- From "CNBC Television" YouTube channel, (link in comment)
1:23
rohanpaul_ai's profile picture

CNBC interviewer asked Palantir CEO Alex Karp how he would defend Wall Street’s concern that AI could replicate what Palantir is doing. Karp defended by basically saying that AI companies may have great engineers, but they do not deeply understand the messy, high-stakes enterprise problems Palantir solves on the ground. ------ Alex Karp: "No one in enterprise factually is worried. I've spent all my life, for better or worse, dealing with the most complicated, most interesting enterprises. I'm on the ground floor of that, probably like no one else. Those kinds of engineers are great engineers, and I'm telling you, they don't talk to the enterprises or understand the technical challenge. If you want to manufacture a car and you need a part, or you want to send a rocket to the moon, or you want to put a missile on your adversary's head and bring America home safely, that stuff doesn't ship. And by the way, there is not a single high-end enterprise like that that would ever put that in place. That is before you even get to the cultural impasse." ---- From "CNBC Television" YouTube channel, (link in comment)

Rohan Paul

97,629 views • 1 day ago

"I don't prompt Claude anymore. I have loops running that prompt Claude and figuring out what to do. My job is to write loops. And this is transition we're going to see for the rest of the year." - Boris Cherny, head of Claude Code at Anthropic.
0:40
rohanpaul_ai's profile picture

"I don't prompt Claude anymore. I have loops running that prompt Claude and figuring out what to do. My job is to write loops. And this is transition we're going to see for the rest of the year." - Boris Cherny, head of Claude Code at Anthropic.

Rohan Paul

2,302,968 views • 29 days ago

Anthropic’s Dario Amodei has only 1 direct report, his chief of staff. The rest of Anthropic’s executive system flows through Dario’s sister, Anthropic President Daniela Amodei, who handles daily operations and reports to the board. For some comparison, OpenAI CEO Sam Altman has around half a dozen direct reports, while Nvidia Corp. CEO Jensen Huang has 60 people reporting to him. --- From "Bloomberg Originals" YouTube channel, (link in comment)
0:28
rohanpaul_ai's profile picture

Anthropic’s Dario Amodei has only 1 direct report, his chief of staff. The rest of Anthropic’s executive system flows through Dario’s sister, Anthropic President Daniela Amodei, who handles daily operations and reports to the board. For some comparison, OpenAI CEO Sam Altman has around half a dozen direct reports, while Nvidia Corp. CEO Jensen Huang has 60 people reporting to him. --- From "Bloomberg Originals" YouTube channel, (link in comment)

Rohan Paul

1,318,360 views • 25 days ago

Jensen Huang on 'Nvidia vs. custom ASIC' He was answering to UBS research analyst question on how custom ASICs will affect NVIDIA or how they are going to compete with custom ASIC. Basically he says - Your custom chip is a science project in a world where NVIDIA is building revenue-generating AI factories. While the competition is still desperately trying to copy Nvidia's last generation, their roadmap is already at the limits of physics. They don't just sell silicon; they deliver the entire, impossibly complex platform that the industry has already surrendered to and standardized on. When you must bet hundreds of billions on your company's future, there is no alternative—they are the only platform to build on, and everyone knows it. --- From 'GTC Financial Analyst Q&A' session (full link in comment)
5:43
rohanpaul_ai's profile picture

Jensen Huang on 'Nvidia vs. custom ASIC' He was answering to UBS research analyst question on how custom ASICs will affect NVIDIA or how they are going to compete with custom ASIC. Basically he says - Your custom chip is a science project in a world where NVIDIA is building revenue-generating AI factories. While the competition is still desperately trying to copy Nvidia's last generation, their roadmap is already at the limits of physics. They don't just sell silicon; they deliver the entire, impossibly complex platform that the industry has already surrendered to and standardized on. When you must bet hundreds of billions on your company's future, there is no alternative—they are the only platform to build on, and everyone knows it. --- From 'GTC Financial Analyst Q&A' session (full link in comment)

Rohan Paul

112,006 views • 2 days ago

Mark Cuban on the next job wave. Customized AI integration for small to mid-sized companies. "Software is dead because everything's gonna be customized to your unique utilization. Who's gonna do it for them... And there are 33 mn companies in the US."
2:25
rohanpaul_ai's profile picture

Mark Cuban on the next job wave. Customized AI integration for small to mid-sized companies. "Software is dead because everything's gonna be customized to your unique utilization. Who's gonna do it for them... And there are 33 mn companies in the US."

Rohan Paul

6,636,178 views • 4 months ago

dot-com bubble vs. a possible AI bubble. From the famous "Dean of Valuation", Professor Aswath Damodaran, of NYU Stern School of Business, “And that’s the real big difference between the dot-com boom and bust and the AI boom. We don’t know whether there’ll be a bust. History suggests there will be a bust. The dot-com boom and bust had no huge capital expenditure in that cycle. In fact, there was very little traditional CapEx, or even R&D, driving it. People started apps. They basically started going on it. This has been the biggest infrastructure run-up I think I’ve ever seen in business. You can go back and compare it to the automobile business 100 years ago. The amount of money that’s being put into AI CapEx is immense, which means that when the correction comes, the pain will be more intense. And herein lies the second problem. The dot-com boom and bust was almost entirely equity-funded. You think, so what? Well, when the bust came, those shareholders lost 60%, 70%, 80%, or 90% of their money. You felt sorry for them, but the loss was restricted to the shareholders. The problem with the AI CapEx boom is that not only is it immense, but a big chunk of it is funded with debt, and the debt is coming from private capital rather than banks. There’s a very real chance that if there’s a correction and companies start having problems, that problem is going to show up as distress and default, and that really doesn’t stay restricted. It spills over into the rest of society. I’m not saying it’s going to be 2008, but 2008 is an example of what happens when lenders overreach, when they lend money at too low a rate, and the correction comes. The pain spills over. So that is my concern with this big market illusion: the potential societal cost of having to deal with debt coming due that you’re unable to pay. It’s much more painful than your share price dropping 90% and you feeling the pain." ---- From "Excess Returns" YouTube channel, (link in comment)
1:40
rohanpaul_ai's profile picture

dot-com bubble vs. a possible AI bubble. From the famous "Dean of Valuation", Professor Aswath Damodaran, of NYU Stern School of Business, “And that’s the real big difference between the dot-com boom and bust and the AI boom. We don’t know whether there’ll be a bust. History suggests there will be a bust. The dot-com boom and bust had no huge capital expenditure in that cycle. In fact, there was very little traditional CapEx, or even R&D, driving it. People started apps. They basically started going on it. This has been the biggest infrastructure run-up I think I’ve ever seen in business. You can go back and compare it to the automobile business 100 years ago. The amount of money that’s being put into AI CapEx is immense, which means that when the correction comes, the pain will be more intense. And herein lies the second problem. The dot-com boom and bust was almost entirely equity-funded. You think, so what? Well, when the bust came, those shareholders lost 60%, 70%, 80%, or 90% of their money. You felt sorry for them, but the loss was restricted to the shareholders. The problem with the AI CapEx boom is that not only is it immense, but a big chunk of it is funded with debt, and the debt is coming from private capital rather than banks. There’s a very real chance that if there’s a correction and companies start having problems, that problem is going to show up as distress and default, and that really doesn’t stay restricted. It spills over into the rest of society. I’m not saying it’s going to be 2008, but 2008 is an example of what happens when lenders overreach, when they lend money at too low a rate, and the correction comes. The pain spills over. So that is my concern with this big market illusion: the potential societal cost of having to deal with debt coming due that you’re unable to pay. It’s much more painful than your share price dropping 90% and you feeling the pain." ---- From "Excess Returns" YouTube channel, (link in comment)

Rohan Paul

527,393 views • 16 days ago

Microsoft CEO Satya Nadella's new interivew: Explains how the next AI moat will not be the model you use, but the learning loop only your company can run. He is really asking what happens to the firm when intelligence becomes something you can rent. For a century, companies protected value through people, processes, data, routines, customer memory, and the tacit knowledge buried in daily operations. Foundation models threaten to flatten that advantage because the same general intelligence can be used by everyone. Nadella’s answer is that firms need their own “hill climbing machine,” a private loop where models learn from company-specific tasks, traces, evaluations, and outcomes. That means the real asset is not just the model. The asset is the environment that keeps improving the model in ways competitors cannot copy. Private evals become strategic memory. Workflow traces become training signal. Human judgment becomes a way to steer compounding, not just correct mistakes. This also reframes AI adoption: a company that only consumes a foundation model may gain productivity, but it may leak the deeper value of its operating knowledge. A company that builds a disciplined learning loop can turn everyday work into accumulating IP. The future firm may therefore be measured by how well it converts its unique activity into durable model improvement. The frontier will not belong only to whoever owns the largest model. It will belong to whoever owns the best loop. ---- From "Stanford Online" YouTube channel, (link in comment)
1:30
rohanpaul_ai's profile picture

Microsoft CEO Satya Nadella's new interivew: Explains how the next AI moat will not be the model you use, but the learning loop only your company can run. He is really asking what happens to the firm when intelligence becomes something you can rent. For a century, companies protected value through people, processes, data, routines, customer memory, and the tacit knowledge buried in daily operations. Foundation models threaten to flatten that advantage because the same general intelligence can be used by everyone. Nadella’s answer is that firms need their own “hill climbing machine,” a private loop where models learn from company-specific tasks, traces, evaluations, and outcomes. That means the real asset is not just the model. The asset is the environment that keeps improving the model in ways competitors cannot copy. Private evals become strategic memory. Workflow traces become training signal. Human judgment becomes a way to steer compounding, not just correct mistakes. This also reframes AI adoption: a company that only consumes a foundation model may gain productivity, but it may leak the deeper value of its operating knowledge. A company that builds a disciplined learning loop can turn everyday work into accumulating IP. The future firm may therefore be measured by how well it converts its unique activity into durable model improvement. The frontier will not belong only to whoever owns the largest model. It will belong to whoever owns the best loop. ---- From "Stanford Online" YouTube channel, (link in comment)

Rohan Paul

79,900 views • 3 days ago

Hollywood is on the brink of massive change...
1:39
rohanpaul_ai's profile picture

Hollywood is on the brink of massive change...

Rohan Paul

24,261,176 views • 1 year ago

Aravind Srinivas just explained why China’s open-source AI may become more powerful than ever. And why Anthropic has lobbied very hard for export control. "the only reason why there is even a 12-month gap between open source and frontier models is export controls. But there is a chance that, because of that, they now get really good at the physical layer. One advantage they (China) have is that they can actually build data centers a lot faster. Power is not a problem. Permits are not a problem. People are not a problem. Labor is not a problem. Expertise is not a problem. And so, by forcing them to go out there and build all this, you are converting them into a far more potent competitor." --- From "20VC with Harry Stebbings" YouTube channel (Harry Stebbings ), link in comment
0:57
rohanpaul_ai's profile picture

Aravind Srinivas just explained why China’s open-source AI may become more powerful than ever. And why Anthropic has lobbied very hard for export control. "the only reason why there is even a 12-month gap between open source and frontier models is export controls. But there is a chance that, because of that, they now get really good at the physical layer. One advantage they (China) have is that they can actually build data centers a lot faster. Power is not a problem. Permits are not a problem. People are not a problem. Labor is not a problem. Expertise is not a problem. And so, by forcing them to go out there and build all this, you are converting them into a far more potent competitor." --- From "20VC with Harry Stebbings" YouTube channel (Harry Stebbings ), link in comment

Rohan Paul

135,792 views • 6 days ago

Dario Amodei has a really hardline view that China shouldn’t have strong AI. He says - "That's the national security interest of the US....The case is very clear, the counterarguments against it are just fishy." - From "Dwarkesh Patel" YT channel
0:40
rohanpaul_ai's profile picture

Dario Amodei has a really hardline view that China shouldn’t have strong AI. He says - "That's the national security interest of the US....The case is very clear, the counterarguments against it are just fishy." - From "Dwarkesh Patel" YT channel

Rohan Paul

138,992 views • 6 days ago

A Chinese father posted a clip of his daughter crying as she says goodbye to her AI learning robot that broke. humans and AI are forming real bonds.
1:06
rohanpaul_ai's profile picture

A Chinese father posted a clip of his daughter crying as she says goodbye to her AI learning robot that broke. humans and AI are forming real bonds.

Rohan Paul

2,801,252 views • 4 months ago

Somebody quit his data center job and leased an empty warehouse. filled it with rows of server machines. charges clients between $4-6 K p/month for private isolated deployment. Each client receives dedicated machines to host model. He runs all on vLLM.
1:16
rohanpaul_ai's profile picture

Somebody quit his data center job and leased an empty warehouse. filled it with rows of server machines. charges clients between $4-6 K p/month for private isolated deployment. Each client receives dedicated machines to host model. He runs all on vLLM.

Rohan Paul

127,768 views • 7 days ago

Deep seek interesting prompt.. From Reddit
0:20
rohanpaul_ai's profile picture

Deep seek interesting prompt.. From Reddit

Rohan Paul

12,042,826 views • 1 year ago

Opinion from a former Meta PM. And this is from Aravind Srinivas of Perplexity "China can build data centers a lot faster. Power is not a problem. Permits are not a problem. People are not a problem. Labor is not a problem. Expertise is not a problem."
0:57
rohanpaul_ai's profile picture

Opinion from a former Meta PM. And this is from Aravind Srinivas of Perplexity "China can build data centers a lot faster. Power is not a problem. Permits are not a problem. People are not a problem. Labor is not a problem. Expertise is not a problem."

Rohan Paul

86,789 views • 5 days ago

"Anything that involves information, anything short of shaping atoms, AI can do 50% of all those jobs right now" - Elon Musk He says white-collar work goes first because it only manipulates information, not atoms. If a job is basically typing, clicking, researching, writing, analyzing or teaching on a screen, AI can do it. The shift will lag due to inertia, but competition will force faster adoption when AI-heavy firms outperform AI-light firms. And then humanoid robots will take most blue-collar jobs. Once machines can “shape atoms,” robots will do physical work in factories, logistics, maintenance, construction, and even run data centers. He expects far more humanoid robots than humans, initially scarce then plentiful, with the main bottlenecks being power, cooling, and materials, not software. --- Video from 'Peter H. Diamandis' YT channel (full link in comment)
1:00
rohanpaul_ai's profile picture

"Anything that involves information, anything short of shaping atoms, AI can do 50% of all those jobs right now" - Elon Musk He says white-collar work goes first because it only manipulates information, not atoms. If a job is basically typing, clicking, researching, writing, analyzing or teaching on a screen, AI can do it. The shift will lag due to inertia, but competition will force faster adoption when AI-heavy firms outperform AI-light firms. And then humanoid robots will take most blue-collar jobs. Once machines can “shape atoms,” robots will do physical work in factories, logistics, maintenance, construction, and even run data centers. He expects far more humanoid robots than humans, initially scarce then plentiful, with the main bottlenecks being power, cooling, and materials, not software. --- Video from 'Peter H. Diamandis' YT channel (full link in comment)

Rohan Paul

34,378 views • 2 days ago

Anthropic CEO Dario Amodei : "Software is going to become cheap, maybe essentially free. The premise that you need to amortize a piece of software you build across millions of users, that may start to be false. But at the same time, there are whole jobs, whole careers that we've built for decades that may not be present. And, you know, I think we can deal with it. I think we can adjust to it. But I don't, I don't think there's an awareness at all of what, of what is coming here and the magnitude of it." --- From "The Wall Street Journal" YT channel (link in comment)
0:52
rohanpaul_ai's profile picture

Anthropic CEO Dario Amodei : "Software is going to become cheap, maybe essentially free. The premise that you need to amortize a piece of software you build across millions of users, that may start to be false. But at the same time, there are whole jobs, whole careers that we've built for decades that may not be present. And, you know, I think we can deal with it. I think we can adjust to it. But I don't, I don't think there's an awareness at all of what, of what is coming here and the magnitude of it." --- From "The Wall Street Journal" YT channel (link in comment)

Rohan Paul

797,859 views • 1 month ago

🦿Xpeng showed a humanoid robot called IRON whose movement looked so human that the team literally cut it open on stage to prove it is a machine. IRON uses a bionic body with a flexible spine, synthetic muscles, and soft skin so joints and torso can twist smoothly like a person. The system has 82 degrees of freedom in total with 22 in each hand for fine finger control. Compute runs on 3 custom AI chips rated at 2,250 TOPS (Tera Operations Per Second), which is far above typical laptop neural accelerators, so it can handle vision and motion planning on the robot. The AI stack focuses on turning camera input directly into body movement without routing through text, which reduces lag and makes the gait look natural. Xpeng staged the cut-open demo at AI Day in Guangzhou this week, addressing rumors that a performer was inside by exposing internal actuators, wiring, and cooling. Company materials also mention a large physical-world model and a multi-brain control setup for dialogue, perception, and locomotion, hinting at a path from stage demos to service work. Production is targeted for 2026, so near-term tasks will be limited, but the hardware shows a serious step toward human-scale manipulation.
0:17
rohanpaul_ai's profile picture

🦿Xpeng showed a humanoid robot called IRON whose movement looked so human that the team literally cut it open on stage to prove it is a machine. IRON uses a bionic body with a flexible spine, synthetic muscles, and soft skin so joints and torso can twist smoothly like a person. The system has 82 degrees of freedom in total with 22 in each hand for fine finger control. Compute runs on 3 custom AI chips rated at 2,250 TOPS (Tera Operations Per Second), which is far above typical laptop neural accelerators, so it can handle vision and motion planning on the robot. The AI stack focuses on turning camera input directly into body movement without routing through text, which reduces lag and makes the gait look natural. Xpeng staged the cut-open demo at AI Day in Guangzhou this week, addressing rumors that a performer was inside by exposing internal actuators, wiring, and cooling. Company materials also mention a large physical-world model and a multi-brain control setup for dialogue, perception, and locomotion, hinting at a path from stage demos to service work. Production is targeted for 2026, so near-term tasks will be limited, but the hardware shows a serious step toward human-scale manipulation.

Rohan Paul

3,802,270 views • 8 months ago

Perplexity’s CEO Aravind Srinivas is pointing to a quiet shift in AI use: the valuable user is no longer the average user. A single power user can now consume as much compute as an entire small team. "There are real engineers at Meta and other companies spending around $10 million a year per engineer on these coding tools. There are users in Perplexity Computer, who spends upwards of $10,000 a month. Their business runs using agent loops that are running inside these harnesses. Even internally inside our own company, there are some people who have set up these kinds of multi-agent hierarchies and agent loops that look like their own software architecture. I often ask these people to come explain to the rest of the company, “Hey, what are you doing with these tools? You clearly are consuming them way more than what we thought the average person in the company would do.” --- The old software instinct was to chase a billion people doing small actions. Agentic AI changes that math because one skilled operator can create a stream of machine work that runs all day. ---- From "20VC with Harry Stebbings" YouTube channel ( Harry Stebbings ), link in comment
1:34
rohanpaul_ai's profile picture

Perplexity’s CEO Aravind Srinivas is pointing to a quiet shift in AI use: the valuable user is no longer the average user. A single power user can now consume as much compute as an entire small team. "There are real engineers at Meta and other companies spending around $10 million a year per engineer on these coding tools. There are users in Perplexity Computer, who spends upwards of $10,000 a month. Their business runs using agent loops that are running inside these harnesses. Even internally inside our own company, there are some people who have set up these kinds of multi-agent hierarchies and agent loops that look like their own software architecture. I often ask these people to come explain to the rest of the company, “Hey, what are you doing with these tools? You clearly are consuming them way more than what we thought the average person in the company would do.” --- The old software instinct was to chase a billion people doing small actions. Agentic AI changes that math because one skilled operator can create a stream of machine work that runs all day. ---- From "20VC with Harry Stebbings" YouTube channel ( Harry Stebbings ), link in comment

Rohan Paul

56,324 views • 4 days ago

AI is not going to replace human beings,” said Mike Cannon-Brookes 👨🏼‍💻🧢🇦🇺 speaking to Lukas Biewald on Weights & Biases’s podcast. And it definitely stayed with me. AI & agents are only going to enable us to do more, build more, and help humanity realize its full potential. #TLVCPartner
1:06
rohanpaul_ai's profile picture

AI is not going to replace human beings,” said Mike Cannon-Brookes 👨🏼‍💻🧢🇦🇺 speaking to Lukas Biewald on Weights & Biases’s podcast. And it definitely stayed with me. AI & agents are only going to enable us to do more, build more, and help humanity realize its full potential. #TLVCPartner

Rohan Paul

551,189 views • 1 month ago