Don Valentine said most of board members sleep. 😴

Temu parent PDD's shares soar following 93% revenue growth

Chinese e-commerce giant PDD Holdings' revenue increased 93% in the July-September quarter to 68.84 billion yuan ($9.66 billion), well ahead of Wall Street's consensus, sending the Temu bargain app operator's New York-listed shares 18% higher in early trading Tuesday.

Read more here

Today, Lightspeed is announcing that we’re leading a $35M Series A in Pika Labs, the company that’s redesigning the entire video-making and editing experience with #AI.

In 2023, video is arguably the most valuable form of media. Yet, compared to all other forms of media, the process for creating high quality video remains challenging for the masses. We shoot videos on our smartphones and upload them to YouTube, but making professional grade films or animation remains out of reach for most.

AI is making all of the above easier. And given the medium’s appeal relative to other forms of creativity, the opportunity in eliminating this gap between creation and value is staggering.

Enter Pika, which is redesigning the entire video-making and editing experience with AI. While other platforms are focused on only enabling pros, Pika makes video creation accessible for all creators, regardless of talent. In under 6 months in beta, Pika has empowered more than 500K people to bring their creative vision to life and 30M+ views on TikTok.

Through Pika, anyone can generate short video clips in virtually any creative style within seconds, just by typing a few lines of text. Plus, users can:

* Extend videos to any length
* Modify aspects of any video instantly
* Extend videos to any content or aspect ratio 
* Adjust camera movement

And all of this is accessible through their web app and Discord server.

While Pika is just launching publicly today, the founders were destined to start this company years ago. Both spent their earliest years aspiring to pursue creative careers. For Demi Guo, poetry was her passion. And for Chenlin Meng, the dream was to grow up to become an animator, inspired by film studios like Pixar.

Both founders’ interests would ultimately evolve to another form of creativity: computer programming. Impressive academic and professional achievements would soon follow, including both obtaining Ph.Ds from Stanford, studying under some of the AI industry’s most renowned professors, and helping author foundational papers that would chart the course of generative AI in the years to come.

Given such an impressive collection of foundational experiences, rooted in an early passion for creativity, the Pika team seems destined to change how we all share our stories visually.

At Lightspeed, we couldn’t be more excited to partner with Pika and support their mission to allow anyone to bring their creative vision to life. And we’re thrilled to be investing alongside other amazing investors at the forefront of AI, including Nat Friedman, Daniel Gross, Elad Gil, Adam D'Angelo, Alex Chung, Clem Delangue 🤗, Andrej Karpathy, Craig Kallman, Keith Peiris, Mati Staniszewski, Aravind Srinivas, Ben Tossell, and funds like Conviction, Factorial, Homebrew, SV Angel, and others.

Through Pika, now anyone can become a video creator and bring their vision to life. Head to www dot pika dot art to start creating 🚀

📸 Look at this post on Facebook https://www.forbes.com/sites/kenrickcai/2023/11/27/pika-ai-video-generator-editor-series-a/?ss=ai&sh=416f356c421b

RIP my legend and teacher, Charlie Munger.

Thank you so much.

Note from poor Charlie’s Almanack.

Friendship

"I think a life properly lived is just learn, learn, learn all the time."

— Charlie Munger

딥마인드가 단백질 구조예측, 기상 예측에 이어서 물질 구조예측을 내놨는데 이것도 그래프 네트워크를 활용했네요.

Today at Google DeepMind we’re introducing Graph Networks for Materials Exploration (GNoME) - a computational tool that has already led to the discovery of 2.2 million new crystals. Crystals are essential materials to our everyday lives, from solar cells to computer chips.

Given that roughly 28,000 stable materials were discovered in the last decade, this new discovery is equivalent to nearly 800 years of knowledge!

#GNoME was trained using a technique called ‘active learning’ and helps to predict the stability of the materials, and therefore pinpoint the most usable ones for real-world applications. The work also represents a leap forward in efficiency for this type of model - this could have a significant impact on how much compute is required to discover a new material.

We believe scaling up materials discovery breakthroughs like this could help us tackle some of the major challenges of the 21st century - like finding cleaner forms of producing electricity.  Read in our blog post about what this research could mean for the future of materials science:

https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/

GNoME shows the potential of using AI to discover and develop new materials at scale. External researchers in labs around the world have independently created 736 of these new structures experimentally in concurrent work.

OpenAI 코파운더인 Greg Brockman이 트윗으로 올렸던 내용.
"놀랍게도 GPT-4에서 생성된 영상의학 보고서 요약이 어떤 경우에는 숙련된 영상의학 전문의가 작성한 보고서 요약보다 선호된다는 것을 발견했다."
어쩌면 약간 침소봉대하는 내용일수도 있는데, 어쨌든 MS와 함께 EMNLP 2023에 발표했던 논문에서 그렇다고 하네요. arXiv에는 10월말에 공개했던 내용이기도.... 아마 RSNA에 가신 분들이 많아서 조용히 넘어가는 듯 싶네요.
제목:
Exploring the Boundaries of GPT-4 in Radiology
요약:
최근 일반 도메인 대규모 언어 모델(LLM)의 성공으로 자연어 처리 패러다임이 여러 도메인과 애플리케이션에 걸쳐 통합된 기반 모델로 크게 변화했습니다. 이 백서에서는 지금까지 가장 성능이 뛰어난 LLM인 GPT-4의 성능을 영상의학 보고서의 텍스트 기반 애플리케이션에서 평가하여 최첨단(SOTA) 영상의학 전용 모델과 비교하는 데 중점을 둡니다. 다양한 프롬프트 전략을 탐색하면서 다양한 범위의 일반적인 방사선 작업에서 GPT-4를 평가한 결과, GPT-4가 최신 SOTA 방사선 모델보다 성능이 뛰어나거나 동등한 수준인 것으로 나타났습니다. 제로 샷 프롬프트를 통해 GPT-4는 이미 시간적 문장 유사성 분류(정확도) 및 자연어 추론(F1)에서 방사선학 모델보다 상당한 이득(≈ 10% 절대적 개선)을 얻었습니다. 데이터 세트별 스타일이나 스키마를 학습해야 하는 작업(예: 결과 요약)의 경우, GPT-4는 예제 기반 프롬프트를 통해 개선되고 감독된 SOTA와 일치합니다. 보드 인증 방사선 전문의와 함께 실시한 광범위한 오류 분석에 따르면 GPT-4는 미묘한 도메인 지식이 필요한 복잡한 맥락에서 가끔 발생하는 오류만 제외하고 충분한 수준의 방사선학 지식을 갖추고 있는 것으로 나타났습니다. 결과를 요약하면, GPT-4의 결과물은 기존의 수기로 작성된 인상과 전반적으로 비슷한 수준인 것으로 나타났습니다.
https://www.microsoft.com/en-us/research/blog/gpt-4s-potential-in-shaping-the-future-of-radiology/?fbclid=IwAR0gfYDT9lX4YSjMhnaYU17DqAvQELAM7HwrS1prf4amwBi1PEEs_ZNIqD0

27살에 Varda라는 우주 기업을 공동 창업하고 & 동시에 파운더스 펀드에서 파트너로 일하고 있는 Delian Asparouhov가 30살을 마주하면서 남긴 글.

https://delian.io/thirty

그저께 PIKA 1.0이 나왔는데

오늘은 오픈소스 Video Generation 오픈소스인 Vchitect 라는 프로젝트가공개되었네요. T2V, I2V 두 모델을 공개했습니다.

T2V
- 코드: https://t.co/sT24IvlhD2
- 스페이스: https://t.co/bhF6TDUZ67

I2V
- 코드: https://t.co/IryTCXbiiJ
- 스페이스: https://t.co/joJ2jUMmZB

Video Generation도 이제 치열한 경쟁이 시작되겠군요. 과연.. 현타가 올 기업들이 얼마나 생길지..