Sabitlenmiş Tweet
Bill Sun
1.6K posts
Bill Sun
@BillSun_AI
AI Founder, Stanford MATH PhD; earliest transformer AI researchers.
AGI house at Hillsborough Katılım Şubat 2016
1.2K Takip Edilen19.6K Takipçiler
@HenryYin_ I think the two AI research lineages should all come from model based control in the 1960 control theorist
English
World model" can mean a video that looks like Sora, a robot brain, or a set of abstract embeddings. These are completely different things.
Two AI research lineages quietly merged in the last two years to produce this confusion. New piece on what happened:
moe-capital.com/blog-home/the-…
English
The chart is from Physical intelligence website, as the collaboration between PI and @danfei_xu’s lab at Georgia tech:
pi.website/research/human…
Check out The paper:
pi.website/download/human…
English
I was pretty excited after watching the talk that @DrJimFan recently gave at @sequoia that emphasized the first-person view video data (sensorized human data).
I think this might be the scaling path for Robotics to hit a scaling law with manageable data cost.
The first signal I got conviction that first-person view video data (sensorized human data) might be the scaling path is when I saw @danfei_xu's work that when you pretrain with enough diversity, you can get sensorized human data and Teleops Robots data in the same space
See the last two charts:
TSNE analysis on mean-pooled tokens from the final layer of the VLM backbone. With no pre-training, it is clear that the model has disjoint representations between human and robot data. But as pretraining becomes more diverse, latent overlap increases, which correlates with performance on our generalization tasks.
I wonder whether someone can accurately plot out the scaling law coefficient for this path, and use this scaling prediction to figure out the video hours/diversity needed.
English
English
In the talk I did 6 months ago with Jie Tang, Director of Google Deepmind Robotics, we also talked about sensorized human data as one of the major possibilty of the data scaling, it is pretty cool to see that the robotics field has getting stronger signal on one path that is working.
See the talk at:
billsun.ai/podcast/
youtube.com/watch?v=NLLmII…
YouTube
English
"You're not just a piece of software, you're a comprehensive product that combines software, community, and content. I think only this kind of thing can survive."
Actually following my previous podcast with @indigox, the most interesting view is this. AI native product cannot just be a software product, it has to come with community, and content. Since software is getting so easy to be replicated by LLM coding, you really want to build up your own community.
x.com/Wenzi_WW/statu…
English
@BillSun_AI @sequoia If you believe in scaling, scaling will believe in you!
English
Bill Sun@BillSun_AI
"When could we get LLM to be able to write their own on-demand Harness for new Job types, like Data scientist AI, Quant researcher AI, Material Scientist AI, etc? Could we build digital version of Peter @steipete or Boris @bcherny?" I asked this question at AGI house last Monday during a dinner with lots of cool guests (see the video). Since I have been working on open-sourcing our agent harness for data scientist AI/Quant Researcher/Financial Analyst AI (github.com/galpha-ai/Alph…), I feel thatLLM has enough training materials Based on the existing open-sourced Harness code like the OpenClaw, leaked Claude code repo, etc, @OriolVinyalsML who leads Google Deepmind's research, called this idea Meta-Harness, and his prediction is that this should be doable in a quick short time framework (quarter? 1 year?)； Furthermore, he thinks that if he had to choose what matters for the next phase, his choice is: LLM > harness. I kind of agree: if LLM can write any harness for newer jobs, figure it how to manage context, design memory system like Peter or Boris, then we are one step closer to AGI.
ZXX
"When could we get LLM to be able to write their own on-demand Harness for new Job types, like Data scientist AI, Quant researcher AI, Material Scientist AI, etc? Could we build digital version of Peter @steipete or Boris @bcherny?"
I asked this question at AGI house last Monday during a dinner with lots of cool guests (see the video).
Since I have been working on open-sourcing our agent harness for data scientist AI/Quant Researcher/Financial Analyst AI (github.com/galpha-ai/Alph…), I feel thatLLM has enough training materials Based on the existing open-sourced Harness code like the OpenClaw, leaked Claude code repo, etc,
@OriolVinyalsML who leads Google Deepmind's research, called this idea Meta-Harness, and his prediction is that this should be doable in a quick short time framework (quarter? 1 year?)；
Furthermore, he thinks that if he had to choose what matters for the next phase, his choice is:
LLM > harness.
I kind of agree: if LLM can write any harness for newer jobs, figure it how to manage context, design memory system like Peter or Boris, then we are one step closer to AGI.
English
Claude is launching a financial service MCP to include these financial ability as MCP and plugin, feel free to try out our open-source repo on top of these plugin:
1）Live Trading:
github.com/galpha-ai/trad…
2) Research and Backtest
github.com/galpha-ai/Alph…
x.com/claudeai/statu…
Claude@claudeai
New for financial services: ready-to-run Claude agent templates for building pitches, conducting valuation reviews, closing the books at month-end, and more. Install them as plugins in Cowork and Claude Code, or use our cookbooks to run them in production as Managed Agents.
English
Podcast talk with @indigox
Founder of Chinese version Twitter
Wickey@Wenzi_WW
In today's podcast, we sit down with @indigox, the co-founder of Weibo, China's defining social media platform (Like X in US). He led product and technology from Weibo's founding inside Sina in 2009 through its 2014 IPO. In 2023, he launched a private AI fund in that has backed Anthropic, Cohere, #xAI, TogetherAI, Lambda, and #SpaceX. @BillSun_AI "Right now, we only have a CPU that provides computation, but everything else around it is lacking. Gradually, it will separate out, forming its own independent logic — this is exactly why agents need a lot of things to help them run better." Welcome to the Innovator Coffee, a podcast that bridges the gap between people and the world of AI and innovation. Follow us to explore the top AI products, ecosystem insights, and the emerging trends. Follow us on Sportify, YouTube and Apple to like, share and comment. open.spotify.com/episode/1suQM4…
English
Podcast talk with Founder of Chinese version of X(Twitter)
Wickey@Wenzi_WW
In today's podcast, we sit down with @indigox, the co-founder of Weibo, a social media platform (Like X in US). AI investor who has backed Anthropic, Cohere, #xAI, TogetherAI, Lambda, and #SpaceX. @BillSun_AI "All successful entrepreneurs are very persistent. Look at someone like Elon Musk. Even when the chances of success are extremely low, he can still succeed. That's what you call someone chosen by fate." Welcome to the Innovator Coffee, a podcast that bridges the gap between people and the world of AI and innovation. Follow us on Sportify, YouTube and Apple to like, share and comment. See full session below: youtu.be/KWlFmDMyvVk?si…
English
Here is a Chinese podcast talk #2 with the founder of WeiBo, the Chinese version X(Twitter):
Wickey@Wenzi_WW
In today's podcast, we sit down with indigox, the co-founder of Weibo, a social media platform (Like X in US). AI investor who has backed Anthropic, Cohere, #xAI, TogetherAI, Lambda, and #SpaceX. @BillSun_AI The software has actually died many times. It's been reborn because the architecture has changed. We've changed from a logical architecture to a generative and intelligent one. Welcome to the Innovator Coffee, a podcast that bridges the gap between people and the world of AI and innovation. Follow us on Sportify, YouTube and Apple to like, share and comment. See full session below: youtube.com/watch?v=KWlFmD…
English
Here is a Chinese podcast talk with the founder of WeiBo, the Chinese version X(Twitter):
Wickey@Wenzi_WW
In today's podcast, we sit down with indigox, the co-founder of Weibo, a social media platform (Like X in US). AI investor who has backed Anthropic, Cohere, #xAI, TogetherAI, Lambda, and #SpaceX. @BillSun_AI "You can't replicate an organization like Microsoft, but you can replicate an agent or the structure of a swarm of agents." Welcome to the Innovator Coffee, a podcast that bridges the gap between people and the world of AI and innovation. Follow us on Sportify, YouTube and Apple to like, share and comment. See full session below: youtube.com/watch?v=KWlFmD…
English
