anqi (@DigitalAnqi) header
anqi (@DigitalAnqi) profile picture

Art Engineer |we create our own reality.

New York, USA
Joined May 2024
Photo by anqi (@DigitalAnqi)Photo by anqi (@DigitalAnqi)Photo by anqi (@DigitalAnqi)Photo by anqi (@DigitalAnqi)Photo by anqi (@DigitalAnqi)Photo by anqi (@DigitalAnqi)Photo by anqi (@DigitalAnqi)
What can you build with AI video in 3 hours? We just hosted our first AI TVC workshop at @UCBerkeley — and it was insane 🔥 In just 3 hours, creators built full TVC campaigns with Mulan AI → for a real local brand: Grégoire Restaurant No templates. No shortcuts. Just workflows. This is what AI video looks like when it’s actually production-ready: 🔹Real creativity 🔹Real execution 🔹Real marketing use cases Huge shoutout to our partner Grégoire & all the talented creators — judging wasn’t easy 👏 ————— 🎬 Mulan AI — the first white-box Chat-to-Workflow AI video agent 🎁 Try MulanAI for free: mulan.pro 🧑‍🤝‍🧑 Join Discord: discord.com/invite/s2XCbHGnC… #MulanAI #AIArtCommuity #ucberkeley #AIworkshop #TVC #Marketing #AItool #Creativity
2
7
1,515
If you are still building video workflows manually, you are out of date. Millions of users on Mulan AI have been creating marketing assets with the first AI content agent who can build workflow and write professional prompts for you. 99% Cost Reduction 90% Efficiency Improvement Full tutorial below (stay for the MV in the end!):
1
2
7
403
who’s gonna be the Kanye of software?
286
636
177
5,090
557,498
An entire AI production house inside your laptop. invideo autonomously takes 500+ creative calls and evals on script, voiceover, sound, and edit to bring ANY idea you have, to life. Created with ❤️ entirely using 'Agents & Models' in invideo. More details coming soon.
84
148
93
840
1,169,588
Every beta user tells us this: Mulan is SO POWERFUL that it’s going to define a brand new era in video realm. Not another model-gathering kit. Not a one-sentence video generator. Not another asset generation / design tool. We go beyond that with editing, covering the full chain of video production. AI Pioneers, do you dare to go beyond the sea with us? 👀 mulan.pro/
1
4
7
433
One click to unlock your entire online fashion store. 👗✨ Upload a photo of a model and a photo of your clothing - #MulanAI turns it into a moving fantasy 🤩 👉Workflow: mulan.pro/flow/k7n4d6 #fashion #AIart #OOTD #ContentMarketing #SmallBusiness
3
11
1
18
2,510
Walking in a dream
🤖 Made with AI
36
99
8
854
35,030
"真正的进步,不在于你记住了多少知识,而在于你的直觉发生了怎样的变化。" ——杨振宁 Intuition is Knowledge Distillation 直觉,就是知识蒸馏 系统二对某类问题的深度思考,经过反复训练,最终被"蒸馏"内化成系统一的快速反应——从费力推理,变成瞬间"看出"答案。 记忆知识只是扩充系统二,而直觉改变是系统一的重构——你对世界的反应方式已在潜意识演化。 真正的学习,就是从系统二到系统一的迁移:从"我会思考"到"我直接知道"。 当思考变成直觉,知识才真正成为你的一部分。
42
296
26
1,204
171,634
The most interesting part for me is where @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 @erikphoel. 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.”
The @karpathy interview 0:00:00 – AGI is still a decade away 0:30:33 – LLM cognitive deficits 0:40:53 – RL is terrible 0:50:26 – How do humans learn? 1:07:13 – AGI will blend into 2% GDP growth 1:18:24 – ASI 1:33:38 – Evolution of intelligence & culture 1:43:43 - Why self driving took so long 1:57:08 - Future of education Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!
225
700
155
5,153
1,053,793
Stop making content. Start creating concepts that get you noticed by top brands. 10 years working with Warner Music, PepsiCo, and other majors taught me: fresh ideas open every door. I built a fresh video concept for you in LTX Studio: 5 brand logos made from… a crowd. Steal my workflow, prompts & action plan to attract big brands with it 👇
103
129
22
1,574
152,313
Recently, n8n announced the beta release of its AI workflow builder. Over two months ago, Mulan AI’s beta users were already able to experience using natural language to automatically build workflows, and we have refined the system four times since then. Repost for beta invite
2
3
3
2,444
👀 Views. Followers. Virality. The currency of the new internet era. Meet Mulan AI — the first video agent built to help creators go viral, grow faster, and never run out of content. ⚡ Coming soon. #GenAI #AIGC #AIVideo #VideoViral #CreativeTools #MulanAI
2
4
435
我看了一下Anthropic分享的Prompt101 其中有两个我觉得可以分享,强化一下使用: 第一个叫做预填充响应 (Pre-fill Response) >做什么: 在API调用中,直接为模型的回答提供一个开头。 >为什么: 这是强制模型输出特定结构(尤其是JSON)的有效技巧。 >示例: 在API请求的“助手(assistant)”部分,直接填入一个左花括号 {,以引导模型续写一个完整的JSON对象。 这个预填充更容易引导模型按照你的输出模式进行响应,因为LLM是文字接龙 第二个是Prompt调试,利用“扩展思考”功能 (Extended Thinking / Scratchpad) >作用: 这是一个强大的调试工具。开启此功能后,模型会显式地展示其“思考过程”或“草稿板”。 >如何使用: 通过分析模型的思考记录,你可以理解它在哪些环节遇到了困难或产生了误解。然后,你可以将这些洞察转化为更清晰的分步指令,固化到你的提示词中,从而使提示词本身变得更强大、更高效。 这个其实是23年宇龙分享过的一个思路,就是当提示词输出效果不符合预期时,需要与模型沟通,确认模型对提示词中的哪部分理解出现了语义偏差。 打印思考也是这种思路的一个变体。 以上是我学到的,视频可能还有其他信息可供学习👇
4
22
122
19,368
15 rules of vibe coding with Cursor
203
710
108
10,299
1,069,090