Rohan Paul (@rohanpaul_ai)
@rohanpaul_ai
Compiling in real-time, the race towards AGI.
The Largest Show on X for AI.
🗞️ Get my daily AI analysis newsletter to your email 👉
6.9K 正在关注 153.4K 粉丝
Jensen Huang on "distillation"
On his new interview with axios, he was asked this question
"Should open source model companies be allowed to distill closed models"
"Distillation—learning from AI, learning from other people, and learning from other sources of knowledge, is fundamental to intelligence.
We are constantly learning from other people. I am learning from you through the questions you are asking, and you are learning from me. All day long, we are learning from one another. AI also has to learn from something.
The original AI models, whether they were open or closed, were trained on previously created knowledge from the internet. Now, AI is generating more content than humans. In a few more years, the internet could be 99% AI-generated content, and that content will have been created by some form of AI.
As a result, AI systems will constantly be distilling knowledge and intelligence from other AI systems. The fact that AI can learn is a good thing. We want AI systems to be intelligent because a smarter AI can also be a safer AI."
----
From "Axios" YouTube channel, (full video link in comment)
显示更多
This is Moonshot AI/Kimi K3's founder & CEO Yang Zhilin at GTC 2026
The 34-year-old researcher studied at Tsinghua, then earned his Carnegie Mellon Ph.D.
Then he worked with Jason Weston at Meta AI and Quoc Le at Google Brain. His work also connected him to Yoshua Bengio and Yann LeCun on the topics of reasoning and representation.
Eventually, Yang returned to China and co-founded Moonshot AI in Beijing a few years later.
His earlier Transformer-XL and XLNet work addressed how language models forget distant context. Moonshot AI carried that research focus forward through Kimi K2’s 1T-parameter mixture-of-experts design.
XLNet, was later cited more than 10,000 times and is still his best-known academic paper.
Kimi K3 expands the formula with native vision, 1M-token context, and only 16 active experts.
His career has become a textbook example of how quickly global research networks can reshape the way countries compete in AI.
----
From "Kimi AI" YouTube channel, (full video link in comment)
显示更多
MIT just built an AI that can control your body.
It can move your fingers, make you play piano, even if you don’t know the song!
AI decides the hand movement. Wrist pads send signals to your muscles, so your fingers move even if you don’t know how
显示更多
Per The Information, Zhipu AI is also (after DeepSeek) exploring a custom ASIC after GLM-5.2 usage reportedly jumped 27x in one week.
A custom ASIC removes flexibility, but it can cut power draw and per-token cost.
Nvidia GPUs are strong general-purpose machines, but inference at scale has different economics. A fixed model can run better on silicon designed around its own repeated operations.
Zhipu has not chosen a partner, and the project may take more than 2 years.
The pattern is now bigger than one Chinese lab or one model launch. Chinese AI companies are trying to make software, hardware, and deployment less separable.
显示更多
DeepSeek is building an inference chip to cut dependence on Nvidia and Huawei in China’s $50B AI-chip market.
DeepSeek’s chip work is still early, with outside partners and private hiring of chip-design engineers.
The hard part is not drawing a chip, but making it at scale. Advanced foundries and high-bandwidth memory remain chokepoints because U.S. rules restrict Chinese access.
DeepSeek can still gain from a narrower chip built mainly for its own models. A custom inference chip could lower serving costs, reduce power needs, and tighten software-hardware control.
---
reuters .com/world/china/chinas-deepseek-developing-its-own-ai-chip-sources-say-2026-07-07/
显示更多
DeepSeek is building an inference chip to cut dependence on Nvidia and Huawei in China’s $50B AI-chip market.
DeepSeek’s chip work is still early, with outside partners and private hiring of chip-design engineers.
The hard part is not drawing a chip, but making it at scale. Advanced foundries and high-bandwidth memory remain chokepoints because U.S. rules restrict Chinese access.
DeepSeek can still gain from a narrower chip built mainly for its own models. A custom inference chip could lower serving costs, reduce power needs, and tighten software-hardware control.
---
reuters .com/world/china/chinas-deepseek-developing-its-own-ai-chip-sources-say-2026-07-07/
显示更多
More good news for local LLMs.
Tencent’s new Hunyuan Hy3 reaches Gemini 3.5-level physics quality for 35x less cost.
Test was done on atomic[.]chat, a desktop app that runs LLMs locally.
The prompt asked 4 models to build bowling, air hockey, and pool simulations. The harder part was preserving physical cause and effect.
A strike needs collision timing, mass transfer, pin rotation, friction, and believable scattering.
A pool break exposes the same weakness, because every wrong angle compounds immediately.
Interestingly, DeepSeek-V4 spent the highest number of tokens (50,600 ), yet produced the weakest visual physics in this test.
显示更多
New Hunyuan Hy3 hits Gemini 3.5 quality on physics for 35x cheaper!
We gave 4 models the same prompt: build three self-contained HTML5 canvas scenes with real physics demos
Prompts:
- A bowling ball knocking down the pins
- An air hockey rally that ends in a goal
- A pool break scattering the rack
Outputs:
Hunyuan Hy3: 29,757 tokens, $0.006
Gemini 3.5: 23,300 tokens, $0.21
GLM-5.2: 25,454 tokens, $0.07
DeepSeek-V4: 50,600 tokens, $0.009
Tencent's Hy3 matched Gemini across all three: clean collisions, the puck bounced true, the pins scattered like a real strike, the rack broke with real momentum, nothing clipped or floated. GLM is genuinely strong on pure coding tasks, but the moment the job steps outside clean code it gives way. DeepSeek was the letdown, it burned the most tokens of anyone (50k, almost 2x Hy3) and still turned in the weakest scenes
显示更多
Beautiful paper from Google DeepMind.
Explains the pathways from AGI to ASI, and why that jump could happen through several routes.
The authors frame the AGI-to-ASI transition around 4 technical pathways:
- continued scaling of compute, model size, data, and test-time inference;
- algorithmic paradigm shifts beyond today’s transformer-based foundation-model stack;
- recursive self-improvement, where AI accelerates AI R&D and improves future systems; and
- multi-agent collective intelligence, where large populations of specialized agents coordinate into a superhuman group agent.
Scaling may work for a while, but it could hit limits in data, compute, energy, or weaker returns from making systems larger.
Recursive improvement is the most uncertain path, because AI could speed up AI research, but that loop may also slow if hard research problems need real-world testing, scarce hardware, or new ideas.
Multi-agent collectives may be the most underappreciated path, because a society of competent digital workers could outperform a brilliant individual model through specialization, speed, and coordination.
The big point is that ASI may not arrive as 1 sudden event, but as a chain of faster changes as AI helps create better AI and stronger scientific tools.
----
– arxiv. org/abs/2606.12683
Title: "From AGI to ASI"
显示更多
Bloomberg: AI is pushing animation toward a cost collapse, with Hollywood filmmakers saying feature production could fall by 90% as studios, workers, and unions brace for fewer human jobs.
Hollywood employment has already been pressured by strikes, fewer productions, mergers, and AI-driven restructuring.
---
bloomberg .com/news/newsletters/2026-06-28/animation-s-ai-reckoning-filmmakers-say-they-can-make-movies-for-90-less
显示更多
A Japanese dev spotted the trick: ask Claude Code to automatically Find Skills.
Can match your goal to the right tool, using Vercel’s skills CLI across Claude, Codex, Cursor, and Gemini.
so install skill like dev tools rather than rewritten by hand
显示更多
New MIT study. Code volume surges by 300%, but output increases by only 30%: The AI dividend meets an awkward reality
Autonomous AI coding agents raised commits by 180%, but releases rose only 30%.
The paper’s main idea is that software production has weak links, so faster code writing does not help as much when humans still need to review, connect, test, package, and ship the work.
The authors also check app marketplaces and find more new apps, but no increase in total usage, which means more software appeared without clear evidence that users adopted more software.
The marketplace evidence points the same way: more new apps appeared, but total usage did not rise.
The authors compare more than 100,000 GitHub developers before and after they start using 3 generations of AI coding tools, from autocomplete to more independent coding agents.
Autocomplete raised commits by 40%, interactive coding agents raised them by 140%, and autonomous coding agents raised them by 180%.
The 180% commit gain shrank to 50% for the number of projects and 30% for actual releases.
The estimated "elasticity of substitution" is 0.25 i.e. for every big improvement in AI’s usefulness, only a small amount of human work can be replaced.
Because AI can write code faster, but humans are still needed to decide what to build, check if the code works, connect it with the rest of the product, fix messy edge cases, and actually ship it.
---
papers .ssrn.com/sol3/papers.cfm?abstract_id=6859839
显示更多
Nemotron 3 Ultra vs GPT-5.5 on atomic[.]chat, a desktop app that runs LLMs locally.
Nemotron 3 Ultra gave almost similar result on a test to build HTML5 canvas with real physics, while being 10X cheaper.
- Nemotron 3 Ultra: 11.3k tokens, $0.051
- GPT 5.5: 11.0k tokens, $0.57
Nemotron 3 Ultra has 550 bn total parameters (55 bn active per token), because it is a Mixture-of-Experts model.
显示更多
Nemotron 3 Ultra performed GPT 5.5 level 10× cheaper
We gave three same prompts to build HTML5 canvas with real physics. At first scene we have water in a spinning drum. Galton board - balls through pegs into bins. And a block collision setup with extreme mass differences.
Outputs:
Nemotron 3 Ultra: 11.3k tokens, $0.051
GPT 5.5: 11.0k tokens, $0.57
Nemotron stays right on GPT 5.5's heels, but at 10× cheaper. The gap in quality is far smaller than the gap in price.
显示更多
Satya Nadella on Microsoft’s Fairwater data center, an AI superfactory.
at today's Microsoft Build 2026 keynote.
its vertically designed, two-story AI data center architecture.
Instead of spreading compute only across a flat floor, Microsoft can place racks in three dimensions, packing far more GPUs densely while preserving fast network access.
This helps the cluster behave more like one massive AI machine, with low latency and high bandwidth between GPUs.
The other major point is its cooling efficiency: its cooling loop is filled once and can operate with effectively zero ongoing water consumption, using roughly the annual daily-water equivalent of a single restaurant.
----
From "Microsoft" YouTube channel, (link in comment)
显示更多
Great to be back at Microsoft Build today. For us, it is not about any one piece of technology or even the platform.
It is about how we can build a frontier intelligence ecosystem together.
Sharing some of our big announcements today ...
显示更多
Bernie Sanders announces he will soon introduce the American AI Sovereign Wealth Fund Act.
It proposes a one-time 50% tax on stock (not profits) in America’s largest AI companies, giving the public a direct 50% ownership stake. 😅
显示更多
Sam Altman's new interview: AI should not be designed to pursue goals that are disconnected from human needs. People must remain at the center of AI development.
“I have no interest in building a super-smart AI that accomplishes some non-human goals. People should react. People should say, ‘Hey, this is what I want, and this is what I do not want.’
I do not think the issue is that we have failed to explain the benefits. We say, ‘AI is going to cure a bunch of diseases,’ and people say, ‘Okay, that is great, but that is not really my question. My question is: What is my role in the future? What is my economic future? What is my agency? How do I know that my kids and my family will still be able to have fulfilling, creative expression, struggle, drive the world forward, grow, and do this thing together in a way that has worked for a long time?’
When people in AI say, ‘Sure, there are going to be no jobs,’ or ‘50% of jobs are going to go away,’ or ‘90% of jobs are going to go away,’ and ‘AI is going to be smarter than you at everything,’ and ‘We will give you some basic income, but you are not really going to have a role,’ that is horrible.
And by the way, if an AI company says, ‘Maybe we are going to destroy all the jobs, and we will be the most valuable company in the world,’ people should look at you like, ‘Yeah, that is a terrible message.’
I do not think the problem is that we have not articulated the upsides. I think people actually believe us. They hear, ‘AI may cure your cancer,’ and they think, ‘That sounds great.’
I think we, as an industry, have failed to explain how people stay in control of determining the future at every step, and how people can still have a meaningful life in all the ways we care about.”
----
From "CNBC Television" YouTube channel, (link in comment)
显示更多
"Every 10 seconds, global token demand is around 31.7 billion in 2026. By 2030 its 1.27 trillion, a 40x increase."
~ Qualcomm CEO Cristiano Amon:
---
The token explosion is not mainly about smarter answers; it is about AI moving from human-paced interaction to Agent-paced activity.
Once agents become persistent, the economy of AI stops will be the background infrastructure.
Every useful action has a hidden bill: context must be carried, memory must be updated, sensors may need to be interpreted, and mistakes must be caught before they become expensive.
----
From "Reuters" YouTube channel, (link in comment)
显示更多
New video of Qualcomm CEO Cristiano Amon: AI will require “gazillions” of tokens.
Because, Agentic AI will consume dramatically more tokens because it performs autonomous tasks, uses multiple systems, and interacts with tools.
AI demand will grow hugely when software starts letting agents act, not just answer.
A chatbot spends tokens on language; an agent spends tokens on deciding, checking, calling tools, reading outputs, revising plans, and coordinating with other software.
Today a single human-AI exchange may be large, a reasoning task may be much larger, but we are already entering the agentic era, where an autonomous workflow can become exponentially larger still because the model is no longer producing one response.
It is running a process.
When a SaaS product redesigns itself around human-agent interaction, every task can become a chain of hidden micro-decisions, and each micro-decision consumes context, memory, tool calls, verification, and output tokens.
Another point is people hear “more tokens” and think only of bigger data centers, when the deeper change is economic: software usage may stop being measured mainly by clicks, seats, or sessions.
It will be measured by how much machine reasoning/tokens gets spent on behalf of each user.
----
From "Reuters" YouTube channel, (link in comment)
显示更多
New video of Qualcomm CEO Cristiano Amon: AI will require “gazillions” of tokens.
Because, Agentic AI will consume dramatically more tokens because it performs autonomous tasks, uses multiple systems, and interacts with tools.
AI demand will grow hugely when software starts letting agents act, not just answer.
A chatbot spends tokens on language; an agent spends tokens on deciding, checking, calling tools, reading outputs, revising plans, and coordinating with other software.
Today a single human-AI exchange may be large, a reasoning task may be much larger, but we are already entering the agentic era, where an autonomous workflow can become exponentially larger still because the model is no longer producing one response.
It is running a process.
When a SaaS product redesigns itself around human-agent interaction, every task can become a chain of hidden micro-decisions, and each micro-decision consumes context, memory, tool calls, verification, and output tokens.
Another point is people hear “more tokens” and think only of bigger data centers, when the deeper change is economic: software usage may stop being measured mainly by clicks, seats, or sessions.
It will be measured by how much machine reasoning/tokens gets spent on behalf of each user.
----
From "Reuters" YouTube channel, (link in comment)
显示更多
A humanoid robot is useful only when teams can test motion, perception and interaction on real hardware, because simulation often misses friction, balance errors, sensor noise, and messy human environments.
That’s exactly why robots need developer ecosystems.
LUMOS Robotics just launched Project EDGE, a program giving 100 free LUMOS NIX humanoids to selected builders, universities, robotics labs, and creative robotics teams.
显示更多
100 FREE LUMOS NIX: LUMOS Robotics Officially Launches Project EDGE!
LUMOS Robotics Founder & CEO Yu Chao officially introduces Project EDGE — inviting global builders, universities, robotics labs, and creative technologists to explore the future of humanoid robotics together.
To supercharge the global developer community, we are providing 100 complimentary LUMOS NIX robots to selected global partners.
Selected partners will receive:
• Complimentary LUMOS NIX units
• Open SDK access
• Direct technical support
From dynamic motion control to embodied AI applications, the stage is yours.
If a LUMOS NIX showed up at your lab tomorrow, what’s the first thing you’d deploy on it? Let us know below!
Welcome to Project EDGE. Apply now via our website.
#humanoidrobot# #robotics#
显示更多
There’s now a platform that hires AI agents for you from 273,000 skills and keeps them running 24/7 while you sleep
@lobehub just launched something called a Chief Agent Operator. you don’t build agents. you don’t prompt agents. you just say what needs doing, and it finds the right agent, deploys it, and reports back through Slack, Discord, or whatever you already use
In 2015, you hired humans on Fiverr for $5/hr. in 2026, AI is hiring AI like it has a budget and a calendar.
Currently most agent-powered workflow asks you to open separate agents, repeat context, assign tasks, check progress, move results between apps, and decide when to escalate work.
LobeHub has an operator layer that hires agents from a 273K-skill marketplace, schedules them in the cloud 24/7, and sends reports through the IM apps where teams already work.
So their "Task" turns an agent into a background worker: you assign the job once, the agent keeps running, shares progress, moves finished work to Pending Review, and updates its work when you leave comments.
50%+ cost savings and cloud 24/7, no self-hosting
The way I understand LobeHub: Claude Code, Cursor, and Manus are powerful agents/tools, but LobeHub is the operator layer that decides who does what, when, and how the work comes back to you.
🧵 1.
显示更多
Jensen Huang thinks Dario Amodei's prediction of $1T in AI revenue by 2030 is too conservative.
"I believe Dario and Anthropic are going to do way better than that. Way better than that.
And the reason for that is the one part that he hasn't considered: I believe every single enterprise software company will also be a value-added reseller of Anthropic's tokens.
And they’re going to get this logarithmic expansion. Their go-to-market is going to expand tremendously this year."
---
From @theallinpod YT channel (link in comment)
显示更多
Chamath: AI advantage may come less from models than from private inputs.
"When labs can build similar models, the real win comes from one unique ingredient in order to monetize it well.
Here is a basic thing about machine learning that is worth knowing: if you take 1,000 of the same inputs and give them to Facebook, Microsoft, Google, and Amazon, they will all come up with the same machine learning model.
But if you have one extra thing, one little ingredient that all of those other companies do not have, your output can be markedly different.
It is like giving two great chefs three ingredients, but giving the third chef one extra ingredient. That person has the ability to do something very special.
Right now, we are in a world where everybody is crawling the open web. We are going to move to a world where, as everybody gets sophisticated enough and information is widely available, somebody is going to say, “You know what? This site, I am not going to allow anybody else to access. It is only for me, only for my models.” Those models will become better.
So we have to let that play out a little bit. It is going to be a really interesting arms race.
The next wave of M&A, for example, could be companies like Google, Microsoft, and Facebook looking at these companies and saying, “Can they be viable inputs to my large language models or to my other machine learning and AI models?”
---
A company with unique workflows, transactions, medical records, industrial logs, legal archives, design files, or user behavior can turn boring private data into a compounding advantage.
Some startups may never become great public companies on their own, yet still become valuable because they own a data stream that makes a larger AI system sharper, more differentiated, or harder to copy.
That turns acquisition strategy upside down: the buyer may not be purchasing revenue, brand, or even software, but a private ingredient for intelligence.
----
From "iConnections" YouTube channel, (link in comment)
显示更多
Chamath: AI advantage may come less from models than from private inputs.
"When labs can build similar models, the real win comes from one unique ingredient in order to monetize it well.
Here is a basic thing about machine learning that is worth knowing: if you take 1,000 of the same inputs and give them to Facebook, Microsoft, Google, and Amazon, they will all come up with the same machine learning model.
But if you have one extra thing, one little ingredient that all of those other companies do not have, your output can be markedly different.
It is like giving two great chefs three ingredients, but giving the third chef one extra ingredient. That person has the ability to do something very special.
Right now, we are in a world where everybody is crawling the open web. We are going to move to a world where, as everybody gets sophisticated enough and information is widely available, somebody is going to say, “You know what? This site, I am not going to allow anybody else to access. It is only for me, only for my models.” Those models will become better.
So we have to let that play out a little bit. It is going to be a really interesting arms race.
The next wave of M&A, for example, could be companies like Google, Microsoft, and Facebook looking at these companies and saying, “Can they be viable inputs to my large language models or to my other machine learning and AI models?”
---
A company with unique workflows, transactions, medical records, industrial logs, legal archives, design files, or user behavior can turn boring private data into a compounding advantage.
Some startups may never become great public companies on their own, yet still become valuable because they own a data stream that makes a larger AI system sharper, more differentiated, or harder to copy.
That turns acquisition strategy upside down: the buyer may not be purchasing revenue, brand, or even software, but a private ingredient for intelligence.
----
From "iConnections" YouTube channel, (link in comment)
显示更多
