正在加载视频...

视频加载失败

返回首页

I have been playing with various natural-language-to-CAD tools integrated with Claude this past week. They have reached an interesting threshold now that you can see that they are capable of generating geometrically detailed models. It isn't perfect zero-shot generation yet but it feels remarkable that you can just take...

ankurhandos's profile picture

Ankur Handa

8,883 subscribers

30,720 次观看 • 4 个月前 •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Right now, I have a few dozen tools to automate my life. Every single thing that I do more than once every week is now a tool: either a Claude Code skill, a scheduled workflow, or an application. • I spend the time doing this once. • I schedule it. • I forget about it (or at least, try to) CREAO is one of the platforms I've used extensively for building some of these automations. They are partnering with me on this post. They are one of the only platforms where you can go from a conversation to a scheduled agent that quickly. This is how it works: 1. You describe the problem you want to solve 2. CREAO's agent builds the logic and executes it 3. You can iterate with the agent to improve the solution 4. When you are done, you turn that solution into a mini app 5. You can schedule that app to run any time you need it Something important: These agents are deterministic. They don't use an LLM to generate answers, so they will always return the same output given the same input. This is critical: when you're automating something that runs every Monday at 9 am, you need it to return the same result every time, not a "creative" answer.
5:00

Right now, I have a few dozen tools to automate my life. Every single thing that I do more than once every week is now a tool: either a Claude Code skill, a scheduled workflow, or an application. • I spend the time doing this once. • I schedule it. • I forget about it (or at least, try to) CREAO is one of the platforms I've used extensively for building some of these automations. They are partnering with me on this post. They are one of the only platforms where you can go from a conversation to a scheduled agent that quickly. This is how it works: 1. You describe the problem you want to solve 2. CREAO's agent builds the logic and executes it 3. You can iterate with the agent to improve the solution 4. When you are done, you turn that solution into a mini app 5. You can schedule that app to run any time you need it Something important: These agents are deterministic. They don't use an LLM to generate answers, so they will always return the same output given the same input. This is critical: when you're automating something that runs every Monday at 9 am, you need it to return the same result every time, not a "creative" answer.

Santiago

13,712 次观看 • 4 个月前

I asked Garry Tan how to use meta prompting to get better at AI: "My partners at YC Jared Friedman and Pete Koomen showed me how to do this. You can take almost anything that you do all the time and just drop it into a context window. And then say, “Here’s a bunch of inputs and outputs." And maybe you also add a bunch of notes. And then you tell it, “Write me a prompt that can act as an agent that takes this input and makes this output over here.” You can do this for almost any type of knowledge work. And you can even introspect. "What are things you notice that I did to convert this from the input to the output?”. And then you can just start using the prompt. Initially, it’s going to suck. Because it’s just not that smart yet. But what’s funny is now, I also use it to Iterate my writing. You can be very direct, "I would never say that", "Don’t say it like this", or "Oh, you used the long word there, use the short word". Just speak to it conversationally. And then when you're happy with the output, you can use that new output to make a new prompt. "Based on this conversation, give me a better initial prompt that incorporates all the things we talked about." And you can do this with literally everything. And in theory, there’s so much it applies to that people do day-to-day. You could use it for tweets. You could use it for editing podcasts. You can use it for pretty much everything. I have a folder of prompts that I use all the time. My YouTube prompt is on v27 or something. I'll go through this process with all the different max models. I'll use GPT 5.2 Pro. I’ll use Grok. I'll use Claude. Then, I’ll take all the outputs from all the models and put them into Claude and say "Here’s my prompt, here’s the output from four LLMs, including yourself. Rate each response and tell me what the pros and cons of each approach are." And I usually say "give it to me in numbered form". And then you can agree with one, disagree with two, tell it three is this or that. And then after that, you say given all of this, synthesize it."
4:16

I asked Garry Tan how to use meta prompting to get better at AI: "My partners at YC Jared Friedman and Pete Koomen showed me how to do this. You can take almost anything that you do all the time and just drop it into a context window. And then say, “Here’s a bunch of inputs and outputs." And maybe you also add a bunch of notes. And then you tell it, “Write me a prompt that can act as an agent that takes this input and makes this output over here.” You can do this for almost any type of knowledge work. And you can even introspect. "What are things you notice that I did to convert this from the input to the output?”. And then you can just start using the prompt. Initially, it’s going to suck. Because it’s just not that smart yet. But what’s funny is now, I also use it to Iterate my writing. You can be very direct, "I would never say that", "Don’t say it like this", or "Oh, you used the long word there, use the short word". Just speak to it conversationally. And then when you're happy with the output, you can use that new output to make a new prompt. "Based on this conversation, give me a better initial prompt that incorporates all the things we talked about." And you can do this with literally everything. And in theory, there’s so much it applies to that people do day-to-day. You could use it for tweets. You could use it for editing podcasts. You can use it for pretty much everything. I have a folder of prompts that I use all the time. My YouTube prompt is on v27 or something. I'll go through this process with all the different max models. I'll use GPT 5.2 Pro. I’ll use Grok. I'll use Claude. Then, I’ll take all the outputs from all the models and put them into Claude and say "Here’s my prompt, here’s the output from four LLMs, including yourself. Rate each response and tell me what the pros and cons of each approach are." And I usually say "give it to me in numbered form". And then you can agree with one, disagree with two, tell it three is this or that. And then after that, you say given all of this, synthesize it."

The Peel

51,632 次观看 • 6 个月前

You can just tell Claude Code to learn skills and it will. I asked Claude to create an app that uses tldraw and it one shotted it with sqlite db, and then I asked Claude Code to create a skill so that it could read and write on the canvas. 10 minutes later it could do it. Weirrrrdddd times.
0:25

You can just tell Claude Code to learn skills and it will. I asked Claude to create an app that uses tldraw and it one shotted it with sqlite db, and then I asked Claude Code to create a skill so that it could read and write on the canvas. 10 minutes later it could do it. Weirrrrdddd times.

Riley Brown

73,418 次观看 • 8 个月前

Been using the new Claude 3.5 model as a tester and now that it is out, I can say it is very very impressive, and the "artifacts" that it generates are like a simpler version of Code Interpreter This is a real-time video of me creating a playable game and editing it with Claude
1:51

Been using the new Claude 3.5 model as a tester and now that it is out, I can say it is very very impressive, and the "artifacts" that it generates are like a simpler version of Code Interpreter This is a real-time video of me creating a playable game and editing it with Claude

Ethan Mollick

162,293 次观看 • 2 年前

Adam Carolla was on Joe Rogan’s podcast. Here I’ll paraphrase one of Carolla’s most important points from this discussion: “I realized that you have a skill set. Like martial arts. You know it, you’re comfortable with it, you’re secure in it. You know your abilities there, just like you know your abilities as a comedian or as an archer. For me, I’m a carpenter. I have a skill, a trade. There are things I know well, so I don’t feel insecure. I know what I’m good at, and I’m grounded in that. But a lot of people don’t have that. They don’t have a trade or a skill or anything they can honestly call expertise. If you asked them, ‘What are you an expert at?’ they wouldn’t have an answer. You could name several — UFC, mixed martial arts, jiu-jitsu, podcasting, standup, whatever it is. You could learn another language or master an instrument. But so many people never find that thing. They never develop a skill they can own. And because of that, they walk around in this heightened state of insecurity.” This is an insightful point: once you start getting good at something, and it can be almost anything, you can build from there. You now have something to talk about, something you know more about than most people. And if you can make it even mildly interesting, you can speak about it at length. But many people spend their free time on social media or binge-watching TV. It is strange now: if you talk to younger guys and ask what they’re interested in, they’ll tell you about a Netflix show, or getting high, ordering Uber Eats, binge-watching, and maybe sports gambling. If your life is dull and uneventful, and most of your free time is spent scrolling social media, it will be hard to hold anyone’s interest.
2:08

Adam Carolla was on Joe Rogan’s podcast. Here I’ll paraphrase one of Carolla’s most important points from this discussion: “I realized that you have a skill set. Like martial arts. You know it, you’re comfortable with it, you’re secure in it. You know your abilities there, just like you know your abilities as a comedian or as an archer. For me, I’m a carpenter. I have a skill, a trade. There are things I know well, so I don’t feel insecure. I know what I’m good at, and I’m grounded in that. But a lot of people don’t have that. They don’t have a trade or a skill or anything they can honestly call expertise. If you asked them, ‘What are you an expert at?’ they wouldn’t have an answer. You could name several — UFC, mixed martial arts, jiu-jitsu, podcasting, standup, whatever it is. You could learn another language or master an instrument. But so many people never find that thing. They never develop a skill they can own. And because of that, they walk around in this heightened state of insecurity.” This is an insightful point: once you start getting good at something, and it can be almost anything, you can build from there. You now have something to talk about, something you know more about than most people. And if you can make it even mildly interesting, you can speak about it at length. But many people spend their free time on social media or binge-watching TV. It is strange now: if you talk to younger guys and ask what they’re interested in, they’ll tell you about a Netflix show, or getting high, ordering Uber Eats, binge-watching, and maybe sports gambling. If your life is dull and uneventful, and most of your free time is spent scrolling social media, it will be hard to hold anyone’s interest.

Rob Henderson

191,435 次观看 • 7 个月前

With Google Flow Tools, you can now use natural language to create bespoke tools and workflows. In this video, you'll see how you can: ➡️ Explore and try the various Tools within Google Flow ➡️ Remix a tool to make it even more relevant for your unique workflow ➡️Create your own tool from scratch with a natural language prompt ➡️ Share your tools with a friend for them to try out Whether you’re looking for an image editor, video resizer or custom shaders, you can now create them in Google Flow, no coding experience required.
2:30

With Google Flow Tools, you can now use natural language to create bespoke tools and workflows. In this video, you'll see how you can: ➡️ Explore and try the various Tools within Google Flow ➡️ Remix a tool to make it even more relevant for your unique workflow ➡️Create your own tool from scratch with a natural language prompt ➡️ Share your tools with a friend for them to try out Whether you’re looking for an image editor, video resizer or custom shaders, you can now create them in Google Flow, no coding experience required.

Google Flow

20,037 次观看 • 2 个月前

I've started to understand why I love mecha shows so much recently, they always try to do something new. While some shows may have similar plots, when you get to the heart of their stories, you realize how unique and different they are. Majestic Prince starts with a bang and ends with a bang. It's silly, it's fun, and it can feel like it doesn't take itself too seriously, but trust me, it does. It's not perfect, but it's perfect in its own way. I just wish it wasn't so short, because I love Team Rabbits and the entire cast of supporting characters. They all feel so real and alive. It feels kind of unfair that this franchise is supposed to end with this movie. It wasn't a long series, but it was still a lot of fun for me. I'm going to miss this show. I loved it.
2:40

I've started to understand why I love mecha shows so much recently, they always try to do something new. While some shows may have similar plots, when you get to the heart of their stories, you realize how unique and different they are. Majestic Prince starts with a bang and ends with a bang. It's silly, it's fun, and it can feel like it doesn't take itself too seriously, but trust me, it does. It's not perfect, but it's perfect in its own way. I just wish it wasn't so short, because I love Team Rabbits and the entire cast of supporting characters. They all feel so real and alive. It feels kind of unfair that this franchise is supposed to end with this movie. It wasn't a long series, but it was still a lot of fun for me. I'm going to miss this show. I loved it.

HIM'EL 🗡️ 勇者 #ThankYouMiura

43,573 次观看 • 9 个月前

Broadcom's CEO just told investors that one senior engineer with Claude Opus can do in one week what ten engineers earning $300,000 a year take three months to produce. Every engineering org in the world is running the wrong headcount math. This was not a prediction. It was a report from inside Broadcom. Hock Tan said his engineers are already using Claude Opus for engineering design and code assistance. Not experimenting with it. Using it. And the output they are getting from a single senior engineer in one week is equivalent to what a team of ten would produce over three months. Do that math slowly. Ten engineers at $300,000 a year is $3 million in annual salary. Three months of their time is $750,000 of labor. One engineer with Claude Opus produces the same result in one week. He also said something that most people in the room let slide past them. The engineers who get good at these tools do not just get faster. They get better. The designs they produce with AI assistance are probably better than what they would produce without it. That is the part nobody wants to say out loud. It was never just about speed. The ceiling on what a single engineer can produce just moved. And the teams that have not figured that out yet are spending $750,000 to do what one person with the right tools can do for a fraction of that. What do you guys think about this? (Watch the full interview at Bloomberg Live on YouTube)
1:15

Broadcom's CEO just told investors that one senior engineer with Claude Opus can do in one week what ten engineers earning $300,000 a year take three months to produce. Every engineering org in the world is running the wrong headcount math. This was not a prediction. It was a report from inside Broadcom. Hock Tan said his engineers are already using Claude Opus for engineering design and code assistance. Not experimenting with it. Using it. And the output they are getting from a single senior engineer in one week is equivalent to what a team of ten would produce over three months. Do that math slowly. Ten engineers at $300,000 a year is $3 million in annual salary. Three months of their time is $750,000 of labor. One engineer with Claude Opus produces the same result in one week. He also said something that most people in the room let slide past them. The engineers who get good at these tools do not just get faster. They get better. The designs they produce with AI assistance are probably better than what they would produce without it. That is the part nobody wants to say out loud. It was never just about speed. The ceiling on what a single engineer can produce just moved. And the teams that have not figured that out yet are spending $750,000 to do what one person with the right tools can do for a fraction of that. What do you guys think about this? (Watch the full interview at Bloomberg Live on YouTube)

Ihtesham Ali

97,145 次观看 • 2 个月前

Pi was built when there were already agent harnesses around. Here’s why Mario Zechner(Mario Zechner), found them suboptimal and built Pi, a minimalist self-modifying agent: #1 - Mario initially was a believer in Claude Code: "I was a believer in Claude code because they were the first that packaged agentic search up in a really compelling package. And at the time that fit my workflow really well. Everything around the LLM was kind of nice and tidy and easy to understand. I was super happy. I was proselytising Claude code." #2 - Reverse engineering Claude Code highlighted the degradation that Mario felt as a user: "I personally like simple tools that are stable and that I can rely on. Even if they have non-deterministic parts, all the deterministic parts should be as stable as possible. That was just not the experience with Claude Code around summer 2025. They would take away your control of the context. They would inject stuff behind your back, which is bad. Then, your workflows stopped working because there's now a system reminder that you don't even see in the UI that would modify the behaviour of the model. They would also do this to the system prompt. I built a little service where I can track the progression or evolution of the system, prompt and tool definitions and, with every release, it was messing with stuff. That just messed with my workflows and I don't appreciate that." #3 - PI was built with an appreciation for simple and reliable tools: "If I commit to a development tool, I want it to be a stable, reliable thing like a hammer. I don't want my hammer to break a different spot every day. That's terrible. We need somebody who goes the full velocity kind of way. But I don't want to work with a tool like that."
2:54

Pi was built when there were already agent harnesses around. Here’s why Mario Zechner(Mario Zechner), found them suboptimal and built Pi, a minimalist self-modifying agent: #1 - Mario initially was a believer in Claude Code: "I was a believer in Claude code because they were the first that packaged agentic search up in a really compelling package. And at the time that fit my workflow really well. Everything around the LLM was kind of nice and tidy and easy to understand. I was super happy. I was proselytising Claude code." #2 - Reverse engineering Claude Code highlighted the degradation that Mario felt as a user: "I personally like simple tools that are stable and that I can rely on. Even if they have non-deterministic parts, all the deterministic parts should be as stable as possible. That was just not the experience with Claude Code around summer 2025. They would take away your control of the context. They would inject stuff behind your back, which is bad. Then, your workflows stopped working because there's now a system reminder that you don't even see in the UI that would modify the behaviour of the model. They would also do this to the system prompt. I built a little service where I can track the progression or evolution of the system, prompt and tool definitions and, with every release, it was messing with stuff. That just messed with my workflows and I don't appreciate that." #3 - PI was built with an appreciation for simple and reliable tools: "If I commit to a development tool, I want it to be a stable, reliable thing like a hammer. I don't want my hammer to break a different spot every day. That's terrible. We need somebody who goes the full velocity kind of way. But I don't want to work with a tool like that."

The Pragmatic Engineer

62,825 次观看 • 3 个月前

Clawdbot creator Peter Steinberger 🦞 says Claude Opus is his favorite model, but OpenAI Codex is the best for coding: "OpenAI is very reliable. For coding, I prefer Codex because it can navigate large codebases. You can prompt and have 95% certainty that it actually works. With Claude Code you need more tricks to get the same." "But character wise, [Opus] behaves so good in a Discord it kind of feels like a human. I've only really experienced that with Opus."
1:47

Clawdbot creator Peter Steinberger 🦞 says Claude Opus is his favorite model, but OpenAI Codex is the best for coding: "OpenAI is very reliable. For coding, I prefer Codex because it can navigate large codebases. You can prompt and have 95% certainty that it actually works. With Claude Code you need more tricks to get the same." "But character wise, [Opus] behaves so good in a Discord it kind of feels like a human. I've only really experienced that with Opus."

TBPN

443,119 次观看 • 7 个月前

.Mollie to Republicans dealing with the legacy media — understand your opponent for who they really are: “If you understand that you're not just having a debate with an opponent, but your most hostile and most rabid opponent; if you understand that they are propaganda and then you go into it, I think that's one thing." "If you're just treating them like you still have this 1960s view of what the media are, that's when you're going to get in trouble." A-men.
0:38

.Mollie to Republicans dealing with the legacy media — understand your opponent for who they really are: “If you understand that you're not just having a debate with an opponent, but your most hostile and most rabid opponent; if you understand that they are propaganda and then you go into it, I think that's one thing." "If you're just treating them like you still have this 1960s view of what the media are, that's when you're going to get in trouble." A-men.

Western Lensman

151,393 次观看 • 3 个月前

REAL ESTATE PEOPLE WILL HATE HIM FOR THIS. HE BUILT A CLAUDE AGENT THAT TURNS ANY LISTING INTO A SELLABLE VIDEO ON ITS OWN Playbook: connect Claude to a video generator, paste a listing, get a cinematic tour of every room, sell it to the agent But typing the prompt for every listing doesn't scale. He turned it into a skill his Claude runs on its own Here's how to build the automated version: 1. Connect the video engine once. In Claude, go to Customize, Connectors, Add Custom Connector, name it Higgsfield, and paste the server URL from higgsfield. ai/mcp. Authenticate through your account. No API keys. Now Claude can generate video straight from chat 2. Turn the workflow into a skill. Instead of pasting the same prompt every time, have Claude build a skill. Tell it: "Create a skill called listing-to-video. When I give it a listing URL, scrape the room photos, generate a cinematic clip of each room with Higgsfield, and save them to a folder." Now the whole process is one command, not a wall of text 3. Let the agent run the listing. Hand it a URL and say "run listing-to-video on this." It pulls the photos, fires each room through the video model, and brings the clips back. You wrote the prompt once, inside the skill. You never write it again 4. Stitch and deliver. Drop the clips together into one tour. Send a free sample to the listing's agent, then charge per video or a monthly rate for ongoing listings 5. Scale it with your team. Add a skill that drafts the outreach email and one that builds a simple landing page for the agent. Now one operator runs sourcing, production, and pitching from a single Claude session The edge isn't generating one video. It's building the skill once so every future listing runs itself Bookmark this
0:42

REAL ESTATE PEOPLE WILL HATE HIM FOR THIS. HE BUILT A CLAUDE AGENT THAT TURNS ANY LISTING INTO A SELLABLE VIDEO ON ITS OWN Playbook: connect Claude to a video generator, paste a listing, get a cinematic tour of every room, sell it to the agent But typing the prompt for every listing doesn't scale. He turned it into a skill his Claude runs on its own Here's how to build the automated version: 1. Connect the video engine once. In Claude, go to Customize, Connectors, Add Custom Connector, name it Higgsfield, and paste the server URL from higgsfield. ai/mcp. Authenticate through your account. No API keys. Now Claude can generate video straight from chat 2. Turn the workflow into a skill. Instead of pasting the same prompt every time, have Claude build a skill. Tell it: "Create a skill called listing-to-video. When I give it a listing URL, scrape the room photos, generate a cinematic clip of each room with Higgsfield, and save them to a folder." Now the whole process is one command, not a wall of text 3. Let the agent run the listing. Hand it a URL and say "run listing-to-video on this." It pulls the photos, fires each room through the video model, and brings the clips back. You wrote the prompt once, inside the skill. You never write it again 4. Stitch and deliver. Drop the clips together into one tour. Send a free sample to the listing's agent, then charge per video or a monthly rate for ongoing listings 5. Scale it with your team. Add a skill that drafts the outreach email and one that builds a simple landing page for the agent. Now one operator runs sourcing, production, and pitching from a single Claude session The edge isn't generating one video. It's building the skill once so every future listing runs itself Bookmark this

Yarchi

54,840 次观看 • 2 个月前

Reporter: In your letter to Congress, you said that hostilities with Iran have been terminated. How can you say that with the naval blockade still in place? U.S. President Donald Trump: "Well, it’s a very friendly blockade. Nobody’s even challenging it. Nobody at all is challenging. Just so you understand, many presidents have been involved in things that are very big. They never had to go through anything with respect to Congress. They considered it to be totally unconstitutional. And the Democrats in Congress that pushed this, and probably a couple of Republicans, they’re hurting our negotiating ability, which is infinite right now. And they shouldn’t be doing it. It’s never happened before. No other president’s done it. And I’m not going to be the first."
0:46

Reporter: In your letter to Congress, you said that hostilities with Iran have been terminated. How can you say that with the naval blockade still in place? U.S. President Donald Trump: "Well, it’s a very friendly blockade. Nobody’s even challenging it. Nobody at all is challenging. Just so you understand, many presidents have been involved in things that are very big. They never had to go through anything with respect to Congress. They considered it to be totally unconstitutional. And the Democrats in Congress that pushed this, and probably a couple of Republicans, they’re hurting our negotiating ability, which is infinite right now. And they shouldn’t be doing it. It’s never happened before. No other president’s done it. And I’m not going to be the first."

Open Source Intel

169,224 次观看 • 3 个月前

I built a Claude Code skill that allows it to generate a deep research report over any collection of complex docs (PDFs, Word, Pptx)….and generate word-level citations and bounding boxes directly back to the source! 📝 Check out “/research-docs”. 1. It parses out text and bounding boxes from every doc with liteparse, in seconds. 2. It then generates a full HTML report of the outputs that let you see word-level citations in each page. Raw Claude obviously has deep research capabilities, but it lacks an audit trail back to the source. This skill gives you a researched report that can be audited by others. Check it out: LiteParse:
1:13

I built a Claude Code skill that allows it to generate a deep research report over any collection of complex docs (PDFs, Word, Pptx)….and generate word-level citations and bounding boxes directly back to the source! 📝 Check out “/research-docs”. 1. It parses out text and bounding boxes from every doc with liteparse, in seconds. 2. It then generates a full HTML report of the outputs that let you see word-level citations in each page. Raw Claude obviously has deep research capabilities, but it lacks an audit trail back to the source. This skill gives you a researched report that can be audited by others. Check it out: LiteParse:

Jerry Liu

77,259 次观看 • 4 个月前

The most interesting part for me is where Andrej Karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across every token in a successful trajectory, upweighting even wrong or irrelevant turns that lead to the right answer. > “Humans don't use reinforcement learning, as I've said before. I think they do something different. Reinforcement learning is a lot worse than the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before is much worse.” So what do humans do instead? > “The book I’m reading is a set of prompts for me to do synthetic data generation. It's by manipulating that information that you actually gain that knowledge. We have no equivalent of that with LLMs; they don't really do that.” > “I'd love to see during pretraining some kind of a stage where the model thinks through the material and tries to reconcile it with what it already knows. There's no equivalent of any of this. This is all research.” Why can’t we just add this training to LLMs today? > “There are very subtle, hard to understand reasons why it's not trivial. If I just give synthetic generation of the model thinking about a book, you look at it and you're like, 'This looks great. Why can't I train on it?' You could try, but the model will actually get much worse if you continue trying.” > “Say we have a chapter of a book and I ask an LLM to think about it. It will give you something that looks very reasonable. But if I ask it 10 times, you'll notice that all of them are the same.” > “You're not getting the richness and the diversity and the entropy from these models as you would get from humans. How do you get synthetic data generation to work despite the collapse and while maintaining the entropy? It is a research problem.” How do humans get around model collapse? > “These analogies are surprisingly good. Humans collapse during the course of their lives. Children haven't overfit yet. They will say stuff that will shock you. Because they're not yet collapsed. But we [adults] are collapsed. We end up revisiting the same thoughts, we end up saying more and more of the same stuff, the learning rates go down, the collapse continues to get worse, and then everything deteriorates.” In fact, there’s an interesting paper arguing that dreaming evolved to assist generalization, and resist overfitting to daily learning - look up The Overfitted Brain by Erik Hoel. I asked Karpathy: Isn’t it interesting that humans learn best at a part of their lives (childhood) whose actual details they completely forget, adults still learn really well but have terrible memory about the particulars of the things they read or watch, and LLMs can memorize arbitrary details about text that no human could but are currently pretty bad at generalization? > “[Fallible human memory] is a feature, not a bug, because it forces you to only learn the generalizable components. LLMs are distracted by all the memory that they have of the pre-trained documents. That's why when I talk about the cognitive core, I actually want to remove the memory. I'd love to have them have less memory so that they have to look things up and they only maintain the algorithms for thought, and the idea of an experiment, and all this cognitive glue for acting.”
7:01

The most interesting part for me is where Andrej Karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across every token in a successful trajectory, upweighting even wrong or irrelevant turns that lead to the right answer. > “Humans don't use reinforcement learning, as I've said before. I think they do something different. Reinforcement learning is a lot worse than the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before is much worse.” So what do humans do instead? > “The book I’m reading is a set of prompts for me to do synthetic data generation. It's by manipulating that information that you actually gain that knowledge. We have no equivalent of that with LLMs; they don't really do that.” > “I'd love to see during pretraining some kind of a stage where the model thinks through the material and tries to reconcile it with what it already knows. There's no equivalent of any of this. This is all research.” Why can’t we just add this training to LLMs today? > “There are very subtle, hard to understand reasons why it's not trivial. If I just give synthetic generation of the model thinking about a book, you look at it and you're like, 'This looks great. Why can't I train on it?' You could try, but the model will actually get much worse if you continue trying.” > “Say we have a chapter of a book and I ask an LLM to think about it. It will give you something that looks very reasonable. But if I ask it 10 times, you'll notice that all of them are the same.” > “You're not getting the richness and the diversity and the entropy from these models as you would get from humans. How do you get synthetic data generation to work despite the collapse and while maintaining the entropy? It is a research problem.” How do humans get around model collapse? > “These analogies are surprisingly good. Humans collapse during the course of their lives. Children haven't overfit yet. They will say stuff that will shock you. Because they're not yet collapsed. But we [adults] are collapsed. We end up revisiting the same thoughts, we end up saying more and more of the same stuff, the learning rates go down, the collapse continues to get worse, and then everything deteriorates.” In fact, there’s an interesting paper arguing that dreaming evolved to assist generalization, and resist overfitting to daily learning - look up The Overfitted Brain by Erik Hoel. I asked Karpathy: Isn’t it interesting that humans learn best at a part of their lives (childhood) whose actual details they completely forget, adults still learn really well but have terrible memory about the particulars of the things they read or watch, and LLMs can memorize arbitrary details about text that no human could but are currently pretty bad at generalization? > “[Fallible human memory] is a feature, not a bug, because it forces you to only learn the generalizable components. LLMs are distracted by all the memory that they have of the pre-trained documents. That's why when I talk about the cognitive core, I actually want to remove the memory. I'd love to have them have less memory so that they have to look things up and they only maintain the algorithms for thought, and the idea of an experiment, and all this cognitive glue for acting.”

Dwarkesh Patel

1,052,015 次观看 • 10 个月前

jihoon wants us to promise to have more fun at treasure concerts, film with your left hand and have fun with your right hand 🤙🏻 #지훈 🐶 you’re asking how yesterday was? yesterday was seriously so fun. it really was fun, but hmmm you guys need to have even more fun.. 🐶 you guys need to have more fun. teumes can def go harder than this, but you’re holding back. you guys can def do more, seriously, you really can but everyone kind of hesitates a little. just go out there and enjoy the atmosphere, got it? ok~? 🐶 i’m not saying don’t film or anything like that. you can film, it’s fine, i don’t mind it, because i think that’s the fans’ freedom. but i mean… film with one hand, your left hand, and use your right hand to hold your lightstick and have fun 🐶 but when both hands are on your phone like this… and i’m right in front of you… why are you looking at me through a camera filter first? you came to see me, so why… when i’m standing right in front of you, are you still only looking at me through your phone? it’s just a little disappointing to me 🐶 i’m not saying anyone is doing something wrong or anything like that, i’m just saying it feels a bit sad. i’m literally right in front of you, so why are you still looking at me inside your phone even when i’m right here? 🐶 and i know you want to take pictures, it’s okay. you can film it, keep it. it’s fine, because it’s not prohibited. in concerts where filming isn’t banned, ofc you can film. you want to take photos. ofc you want to take photos when your favorite singer is right in front of you, ofc you want to film it. i know that, we all know that. you can film it but film with your left hand and have fun with your right hand. let’s all make that promise. seriously, let’s really do it, film with your left hand and have fun with your right hand 🐶 because… when i’m like this, from up on stage when we look out like this, your gazes, your lightsticks, those kinds of things... when i feel like i’m directly interacting with each and every one of you like that.. i get out of breath like i might die, i get dizzy and everything, and my adrenaline gets pushed to the limit. but then if it’s all just iphone 17 pro maxes everywhere… it feels a bit disappointing 🐶 but i really do like that you film. i just feel a little regretful like if you film with one hand and have fun with the other.. like use your right hand properly to have fun, and use your left hand to film and capture everything… that’s how it should be, i guess. that kind of feeling 🐶 i’m not saying this to hear sorry from you guys. that’s not what i mean. i’m not saying you’re doing anything wrong, i just mean let’s all have fun together like that 💬 but oppa, if you’re playing with your right hand, the video in your left hand does get shaky so i just jump around and go all out 🐶 but even that shaking… i think of it as part of the live feeling
2:11

jihoon wants us to promise to have more fun at treasure concerts, film with your left hand and have fun with your right hand 🤙🏻 #지훈 🐶 you’re asking how yesterday was? yesterday was seriously so fun. it really was fun, but hmmm you guys need to have even more fun.. 🐶 you guys need to have more fun. teumes can def go harder than this, but you’re holding back. you guys can def do more, seriously, you really can but everyone kind of hesitates a little. just go out there and enjoy the atmosphere, got it? ok~? 🐶 i’m not saying don’t film or anything like that. you can film, it’s fine, i don’t mind it, because i think that’s the fans’ freedom. but i mean… film with one hand, your left hand, and use your right hand to hold your lightstick and have fun 🐶 but when both hands are on your phone like this… and i’m right in front of you… why are you looking at me through a camera filter first? you came to see me, so why… when i’m standing right in front of you, are you still only looking at me through your phone? it’s just a little disappointing to me 🐶 i’m not saying anyone is doing something wrong or anything like that, i’m just saying it feels a bit sad. i’m literally right in front of you, so why are you still looking at me inside your phone even when i’m right here? 🐶 and i know you want to take pictures, it’s okay. you can film it, keep it. it’s fine, because it’s not prohibited. in concerts where filming isn’t banned, ofc you can film. you want to take photos. ofc you want to take photos when your favorite singer is right in front of you, ofc you want to film it. i know that, we all know that. you can film it but film with your left hand and have fun with your right hand. let’s all make that promise. seriously, let’s really do it, film with your left hand and have fun with your right hand 🐶 because… when i’m like this, from up on stage when we look out like this, your gazes, your lightsticks, those kinds of things... when i feel like i’m directly interacting with each and every one of you like that.. i get out of breath like i might die, i get dizzy and everything, and my adrenaline gets pushed to the limit. but then if it’s all just iphone 17 pro maxes everywhere… it feels a bit disappointing 🐶 but i really do like that you film. i just feel a little regretful like if you film with one hand and have fun with the other.. like use your right hand properly to have fun, and use your left hand to film and capture everything… that’s how it should be, i guess. that kind of feeling 🐶 i’m not saying this to hear sorry from you guys. that’s not what i mean. i’m not saying you’re doing anything wrong, i just mean let’s all have fun together like that 💬 but oppa, if you’re playing with your right hand, the video in your left hand does get shaky so i just jump around and go all out 🐶 but even that shaking… i think of it as part of the live feeling

행복지수 314%

23,355 次观看 • 3 个月前

My Latest AI short was made with Seedance 2.0 the model that has been firing up the internet over the past couple of days. I made this in two and half days and it's an original short that is based on HP Lovecraft's work which is now Public Domain (No Copyright infringement here!). If you are not into period romances then bare with it as it gets a more little action packed in the second half! The model is brilliant at both performance and action, you can now make convincing dialogue and action sequences for AI cinema. With this model we have now reached the stage where we can tell any story we want with AI, the tools are capable. This is only 720p at the moment but I suspect it will be upgraded to 1080p very soon. We stand at the threshold of having broadcast quality tools capable of creating anything we can imagine available to everyone, so lets do new an interesting things with it. Thank you to CapCut for whitelisting me and to Dreamina AI AI for giving me access as part of their CPP. Thanks for watching! 👀
3:29

My Latest AI short was made with Seedance 2.0 the model that has been firing up the internet over the past couple of days. I made this in two and half days and it's an original short that is based on HP Lovecraft's work which is now Public Domain (No Copyright infringement here!). If you are not into period romances then bare with it as it gets a more little action packed in the second half! The model is brilliant at both performance and action, you can now make convincing dialogue and action sequences for AI cinema. With this model we have now reached the stage where we can tell any story we want with AI, the tools are capable. This is only 720p at the moment but I suspect it will be upgraded to 1080p very soon. We stand at the threshold of having broadcast quality tools capable of creating anything we can imagine available to everyone, so lets do new an interesting things with it. Thank you to CapCut for whitelisting me and to Dreamina AI AI for giving me access as part of their CPP. Thanks for watching! 👀

Uncanny Harry AI

76,824 次观看 • 6 个月前

Vision-language AI models have a gaze. And you can steer it! 👀 Redirect just 9% of a model’s attention heads to any region in an image, and the VLM will start describing that region mid-generation. We call them Gaze Heads! Try the demo: 🧵👇
0:21

Vision-language AI models have a gaze. And you can steer it! 👀 Redirect just 9% of a model’s attention heads to any region in an image, and the VLM will start describing that region mid-generation. We call them Gaze Heads! Try the demo: 🧵👇

Rohit Gandikota

51,303 次观看 • 2 个月前

An MTV LA journalist on her encounter with ENHYPEN: -I’ve had experience with many kpop bands and all of them have been wonderful. But what I experienced with ENHYPEN was something very different, I’ll tell you why. The people that are behind the kpop industry are people who follow procedures very rigorously. There’s a lot of control over the questions you can ask, of what you get to say, of how they need to answer. If you didn't know, its a very strict world in that regard. So, usually with this type of interview, there's not a lot of room for you to interact with the talents outside of the cameras. Normally they're sitting down, waiting for someone to tell them the camers are rolling and for questions to be asked. They almost never interact or ask you questions, which could be due to the language barrier. But in this case it wasn't like that. From the moment they came in, they were super friendly, they greeted us all. You could tell they had watched the Norway vs. England game, and we started talking to them about it. We told them “right now, we’re rooting for Norway because England eliminated us”. They laughed and said “Ok we support Noway!” They were very chatty, very nice. We ended the “payaso de rodeo” portion super hyped and happy. We sang “ENHYPEN, hermano ya eres Mexicano (ENHYPEN, brother, you’re already Mexican)” and they were like “Wow, what does that mean?” and when we told them, they asked us what to do if the fans yell that at them, and we told them to say “Viva Mexico!”. There were a lot of interactions that didn’t come out on camera and that almost never happens to me with kpop artists. It doesn’t usually happen, so it surprised me that it was like that. It shows that they, even though they're huge worldwide, are still very humble people, very down to Earth. I am going to my grave regretting that dance segment. I don’t know when my brain decided to go the other side (taught them a slightly incorrect dance) But, either way, we had an amazing time. They had a lot of fun. I couldn't stay until the end of the show, so I found out after that they tried to dance it on the stage. But that they didn't do it right. I’m sorry… it was probably my fault! Anyways, I want to tell you that if you are an ENGENE and you support ENHYPEN, honestly, girl (or boy or them), you get everything… They're really people who are super, super nice. I was terrified, because I was called out once before for not knowing how to pronounce a Korean name, so I spent an entire day listening to a video of how to pronounce them, so something like that doesn't happen again. But honestly, from the moment they walked in, they took my nerves away. They made me feel like they were having a good time- I was having a good time, the team was having a good time, everyone was having a good time! And hopefully when ENGENE gets to see this content, they will be having an incredible time. I dont know when the whole interview will come out, but it'll air on the MTV LA channel. You can watch them there. When it comes out, I'll let you know. Tell me what you think~
3:30

An MTV LA journalist on her encounter with ENHYPEN: -I’ve had experience with many kpop bands and all of them have been wonderful. But what I experienced with ENHYPEN was something very different, I’ll tell you why. The people that are behind the kpop industry are people who follow procedures very rigorously. There’s a lot of control over the questions you can ask, of what you get to say, of how they need to answer. If you didn't know, its a very strict world in that regard. So, usually with this type of interview, there's not a lot of room for you to interact with the talents outside of the cameras. Normally they're sitting down, waiting for someone to tell them the camers are rolling and for questions to be asked. They almost never interact or ask you questions, which could be due to the language barrier. But in this case it wasn't like that. From the moment they came in, they were super friendly, they greeted us all. You could tell they had watched the Norway vs. England game, and we started talking to them about it. We told them “right now, we’re rooting for Norway because England eliminated us”. They laughed and said “Ok we support Noway!” They were very chatty, very nice. We ended the “payaso de rodeo” portion super hyped and happy. We sang “ENHYPEN, hermano ya eres Mexicano (ENHYPEN, brother, you’re already Mexican)” and they were like “Wow, what does that mean?” and when we told them, they asked us what to do if the fans yell that at them, and we told them to say “Viva Mexico!”. There were a lot of interactions that didn’t come out on camera and that almost never happens to me with kpop artists. It doesn’t usually happen, so it surprised me that it was like that. It shows that they, even though they're huge worldwide, are still very humble people, very down to Earth. I am going to my grave regretting that dance segment. I don’t know when my brain decided to go the other side (taught them a slightly incorrect dance) But, either way, we had an amazing time. They had a lot of fun. I couldn't stay until the end of the show, so I found out after that they tried to dance it on the stage. But that they didn't do it right. I’m sorry… it was probably my fault! Anyways, I want to tell you that if you are an ENGENE and you support ENHYPEN, honestly, girl (or boy or them), you get everything… They're really people who are super, super nice. I was terrified, because I was called out once before for not knowing how to pronounce a Korean name, so I spent an entire day listening to a video of how to pronounce them, so something like that doesn't happen again. But honestly, from the moment they walked in, they took my nerves away. They made me feel like they were having a good time- I was having a good time, the team was having a good time, everyone was having a good time! And hopefully when ENGENE gets to see this content, they will be having an incredible time. I dont know when the whole interview will come out, but it'll air on the MTV LA channel. You can watch them there. When it comes out, I'll let you know. Tell me what you think~

AboutEnhypen

102,617 次观看 • 1 个月前