Introducing Sequoia’s latest market map!
Health is so fundamental to our quality of life, but healthcare is rife with administrative inefficiencies. Healthcare expenditures are almost 20% of U.S. GDP.
Disruption in this space has always been difficult, but gen AI is poised to change the healthcare industry – not just driving efficiency, but dramatically improving patient outcomes. We are excited to meet the innovators driving change in this space.
Please give us feedback and let us know what we missed! https://lnkd.in/gmcEQ4d4
Health is so fundamental to our quality of life, but healthcare is rife with administrative inefficiencies. Healthcare expenditures are almost 20% of U.S. GDP.
Disruption in this space has always been difficult, but gen AI is poised to change the healthcare industry – not just driving efficiency, but dramatically improving patient outcomes. We are excited to meet the innovators driving change in this space.
Please give us feedback and let us know what we missed! https://lnkd.in/gmcEQ4d4
lnkd.in
LinkedIn
This link will take you to a page that’s not on LinkedIn
중국의 Summarization 연구하던 연구자들이 화가 많이 났네요. 이제 Summarization 연구는 더 이상 끝났다고 논문을 통해 선언합니다.
Summarization is (Almost) Dead
https://arxiv.org/abs/2309.09558
GPT와 같은 Large Language Model (LLM) 들이 너무나 좋은 성능을 보여주면서, 기존의 많은 자연어 처리 연구 주제들을 무용지물로 바꾸어 버렸습니다.
Summarization은 대표적인 자연어 처리 연구 주제 중에 하나이고, 이를 개선하기 위해 정말 다양한 연구들이 활발히 진행됐었구요.
그런데 이제 그 어떤 방법을 써도 LLM들이 할 수 있는 것에 못 미치는 상황이 되었습니다. LLM 연구자들은 보통 summarization은 일종의 test set일 뿐 실제 이를 개선하기 위한 많은 노력을 하지도 않았는데요.
먼가 산업 혁명 시대 변화와 오버랩되네요.
Summarization is (Almost) Dead
https://arxiv.org/abs/2309.09558
GPT와 같은 Large Language Model (LLM) 들이 너무나 좋은 성능을 보여주면서, 기존의 많은 자연어 처리 연구 주제들을 무용지물로 바꾸어 버렸습니다.
Summarization은 대표적인 자연어 처리 연구 주제 중에 하나이고, 이를 개선하기 위해 정말 다양한 연구들이 활발히 진행됐었구요.
그런데 이제 그 어떤 방법을 써도 LLM들이 할 수 있는 것에 못 미치는 상황이 되었습니다. LLM 연구자들은 보통 summarization은 일종의 test set일 뿐 실제 이를 개선하기 위한 많은 노력을 하지도 않았는데요.
먼가 산업 혁명 시대 변화와 오버랩되네요.
arXiv.org
Summarization is (Almost) Dead
How well can large language models (LLMs) generate summaries? We develop new datasets and conduct human evaluation experiments to evaluate the zero-shot generation capability of LLMs across five...
Is only summarization dead? Or will new type of ai replace traditional computings?
Most people missed one of the most important news of the summer - open-access multi-modal (image + text) model coming out of Hugging Face!
You can optimize it, fine-tune it and customize it for your use-case: https://lnkd.in/eWCGJ3Fn
You can try the fun version of it here: https://lnkd.in/eMQA66bq or the general playground here: https://lnkd.in/epbMhXxX
What will/should the community build with it?
You can optimize it, fine-tune it and customize it for your use-case: https://lnkd.in/eWCGJ3Fn
You can try the fun version of it here: https://lnkd.in/eMQA66bq or the general playground here: https://lnkd.in/epbMhXxX
What will/should the community build with it?
lnkd.in
LinkedIn
This link will take you to a page that’s not on LinkedIn
I sent 53 cold emails to Marc Benioff, and each was individually crafted with different hooks, different personal anecdotes, I A/B tested cold emails to the extreme and finally, it worked.
From the future of SF, the balance of in-office vs work from home, and the future of AI, this is a truly special one. Coming on Monday!
#founder #funding #business #investing #vc #venturecapital #entrepreneur #startup #salesforce
From the future of SF, the balance of in-office vs work from home, and the future of AI, this is a truly special one. Coming on Monday!
#founder #funding #business #investing #vc #venturecapital #entrepreneur #startup #salesforce
How Jeff Bezos produces unique insights:
“Jeff has an uncanny ability to read a narrative and consistently arrive at insights that no one else did, even though we were all reading the same narrative.
After one meeting, I asked him how he was able to do that.
He responded with a simple and useful tip that I have not forgotten:
He assumes each sentence he reads is wrong until he can prove otherwise.
He's challenging the content of the sentence, not the motive of the writer.
Jeff, by the way, was usually among the last to finish reading.”
“Jeff has an uncanny ability to read a narrative and consistently arrive at insights that no one else did, even though we were all reading the same narrative.
After one meeting, I asked him how he was able to do that.
He responded with a simple and useful tip that I have not forgotten:
He assumes each sentence he reads is wrong until he can prove otherwise.
He's challenging the content of the sentence, not the motive of the writer.
Jeff, by the way, was usually among the last to finish reading.”
Earlier today, Cisco announced its intention to acquire Splunk for $28b , a 30% premium to the closing price.
Reviewing the financials, we see Splunk is a very healthy business.
$3.6b in revenues growing at 37% places the business in the top quartile of public software companies. The 78% gross margin is 6 percentage points greater than the public median.
The estimated sales efficiency at 0.6 is top quartile. This means for every sales and marketing dollar invested the company purchases an additional 60 cents of gross profit in the next period. The net income margin of negative 7.6% is in line with most other software companies.
Now let's examine the valuation of the business. before the announcement of the acquisition a company traded at a 4.2x forward revenue multiple. The implied multiple after the M&A is 5.7x, assuming the acquisition price is at $28b.
Our internal model predicts the forward multiple based on similar companies should be about 7.5x. It was a bit overoptimistic with the Klaviyo valuation (predicted $10b ; it opened at $9.3b & actual as of today is $8.5b) but this may be due to the recent downdraft in the market after the Fed indicated a longer period of high-rates & alluded to potentially raising rates once more this year.
The Cisco acquisition continues the trend of take-privates that includes New Relic for $6.5b, Qualtrics for $12.5b, & Software AG for $2.4b - all in the last 6 months.
Reviewing the financials, we see Splunk is a very healthy business.
$3.6b in revenues growing at 37% places the business in the top quartile of public software companies. The 78% gross margin is 6 percentage points greater than the public median.
The estimated sales efficiency at 0.6 is top quartile. This means for every sales and marketing dollar invested the company purchases an additional 60 cents of gross profit in the next period. The net income margin of negative 7.6% is in line with most other software companies.
Now let's examine the valuation of the business. before the announcement of the acquisition a company traded at a 4.2x forward revenue multiple. The implied multiple after the M&A is 5.7x, assuming the acquisition price is at $28b.
Our internal model predicts the forward multiple based on similar companies should be about 7.5x. It was a bit overoptimistic with the Klaviyo valuation (predicted $10b ; it opened at $9.3b & actual as of today is $8.5b) but this may be due to the recent downdraft in the market after the Fed indicated a longer period of high-rates & alluded to potentially raising rates once more this year.
The Cisco acquisition continues the trend of take-privates that includes New Relic for $6.5b, Qualtrics for $12.5b, & Software AG for $2.4b - all in the last 6 months.
Chamath Palihapitiya, the “King of SPACs,” lost his investors more than $12B with his 6 SPAC IPOs. Today, Clover and Akili have 0 enterprise value. Virgin Galactic’s is barely hovering at ~$100M.
If you invested $100 into each of Chamath’s SPACs at the peak of the market in Dec 2021, you’d have lost a whopping 73% of your investment. That’s worse than the S&P 500 (-9%), all SPACS (-32%), bitcoin (-44%), and the memestock GameStop (-54%).
So much for the Warren Buffet of the Reddit age. Amazingly, the real Buffet generated a positive return of 22% during the same period.
The poor performance of SPACs — and of Chamath — is a fantastic demonstration of the destructive power of poorly constructed incentives.
—
SPACs were all the rage during the stock mania of 2020 and 2021. Proponents of SPACs argue that they “democratize” access to private unicorns that generally delay going public because of the laborious IPO process.
SPACs provide a way for any private company to go public quickly. SPAC sponsors first IPO a shell company, typically raising hundreds of millions of proceeds. They then hunt for unicorns to “buy” and take public, merging the shell company with the private company.
For doing all this work, the sponsors are compensated with SPAC founder shares worth roughly 20% of the initial capital raised (eg $40M on $200M raised).
Therein lies in the problem. Sponsors are compensated handsomely regardless of these companies’ long-term performance. They just had to make the target companies sound appealing enough to attract enough investors for enough time to sell their founder shares.
—
I gotta give it to him. Chamath really is a great poker player. He cashed in his chips and profited off these deals to the tune of $750M. But those who believed and went “all-in” weren’t so lucky.
해시태그#Chamath 해시태그#SPAC 해시태그#Investing 해시태그#Markets
If you invested $100 into each of Chamath’s SPACs at the peak of the market in Dec 2021, you’d have lost a whopping 73% of your investment. That’s worse than the S&P 500 (-9%), all SPACS (-32%), bitcoin (-44%), and the memestock GameStop (-54%).
So much for the Warren Buffet of the Reddit age. Amazingly, the real Buffet generated a positive return of 22% during the same period.
The poor performance of SPACs — and of Chamath — is a fantastic demonstration of the destructive power of poorly constructed incentives.
—
SPACs were all the rage during the stock mania of 2020 and 2021. Proponents of SPACs argue that they “democratize” access to private unicorns that generally delay going public because of the laborious IPO process.
SPACs provide a way for any private company to go public quickly. SPAC sponsors first IPO a shell company, typically raising hundreds of millions of proceeds. They then hunt for unicorns to “buy” and take public, merging the shell company with the private company.
For doing all this work, the sponsors are compensated with SPAC founder shares worth roughly 20% of the initial capital raised (eg $40M on $200M raised).
Therein lies in the problem. Sponsors are compensated handsomely regardless of these companies’ long-term performance. They just had to make the target companies sound appealing enough to attract enough investors for enough time to sell their founder shares.
—
I gotta give it to him. Chamath really is a great poker player. He cashed in his chips and profited off these deals to the tune of $750M. But those who believed and went “all-in” weren’t so lucky.
해시태그#Chamath 해시태그#SPAC 해시태그#Investing 해시태그#Markets
제가 10.5일부터 SF에서 짧으면 1달 길면 Thanks Giving 까지 머물 예정인데요 😉️️️️️️ 주로 AI로 창업/투자하는 분들, AI Researcher/Engineer 그리고 AI가 아니더라도 재미있는 문제를 푸는 사람들과 교류할 생각입니다. 뉴욕과 남미 쪽도 다녀올 생각이에요!
SF에 있는 창업팀, 투자자, 빌더, 리서처 중에서 추천해주실만한 사람/회사가 있을까요?~ 직접 아시지 못해도 알려주시는 것만으로도 도움 될 것 같습니다. 콜드콜은 자신 있거든요 🫡️️ @startup_learner으로 DM 주세요.
소개해주셔서 만난 경우는 그 분과 대화하고 이야기했던 내용들을 상세히 공유드리도록 해볼게요!
SF에 있는 창업팀, 투자자, 빌더, 리서처 중에서 추천해주실만한 사람/회사가 있을까요?~ 직접 아시지 못해도 알려주시는 것만으로도 도움 될 것 같습니다. 콜드콜은 자신 있거든요 🫡️️ @startup_learner으로 DM 주세요.
소개해주셔서 만난 경우는 그 분과 대화하고 이야기했던 내용들을 상세히 공유드리도록 해볼게요!
👍4❤1
Continuous Learning_Startup & Investment pinned «제가 10.5일부터 SF에서 짧으면 1달 길면 Thanks Giving 까지 머물 예정인데요 😉️️️️️️ 주로 AI로 창업/투자하는 분들, AI Researcher/Engineer 그리고 AI가 아니더라도 재미있는 문제를 푸는 사람들과 교류할 생각입니다. 뉴욕과 남미 쪽도 다녀올 생각이에요! SF에 있는 창업팀, 투자자, 빌더, 리서처 중에서 추천해주실만한 사람/회사가 있을까요?~ 직접 아시지 못해도 알려주시는 것만으로도 도움 될 것 같습니다. 콜드콜은…»
https://www.sequoiacap.com/article/generative-ai-act-two/
This moment has been decades in the making. Six decades of Moore’s Law have given us the compute horsepower to process exaflops of data. Four decades of the internet (accelerated by COVID) have given us trillions of tokens’ worth of training data. Two decades of mobile and cloud computing have given every human a supercomputer in the palm of our hands. In other words, decades of technological progress have accumulated to create the necessary conditions for generative AI to take flight.
ChatGPT became the fastest-growing application with particularly strong product-market fit among students and developers; Midjourney became our collective creative muse and was reported to have reached hundreds of millions of dollars in revenue with a team of just eleven; and Character popularized AI entertainment and companionship and created the consumer “social” application we craved the most—with users spending two hours on average in-app.
Towards Act Two
These applications are different in nature than the first apps out of the gate. They tend to use foundation models as a piece of a more comprehensive solution rather than the entire solution. They introduce new editing interfaces, making the workflows stickier and the outputs better. They are often multi-modal.
The market is already beginning to transition from “Act 1” to “Act 2.” Examples of companies entering “Act 2” include Harvey, which is building custom LLMs for elite law firms; Glean, which is crawling and indexing our workspaces to make Generative AI more relevant at work; and Character and Ava, which are creating digital companions.
This reflects two important thrusts in the market: Generative AI’s evolution from technology hammer to actual use cases and value, and the increasingly multimodal nature of generative AI applications.
The moats are in the customers, not the data. the data that application companies generate does not create an insurmountable moat, and the next generations of foundation models may very well obliterate any data moats that startups generate. Rather, workflows and user networks seem to be creating more durable sources of competitive advantage.
In short, generative AI’s biggest problem is not finding use cases or demand or distribution, it is proving value. What are you going to use all this infrastructure to do? How is it going to change people’s lives?” The path to building enduring businesses will require fixing the retention problem and generating deep enough value for customers that they stick and become daily active users.
If you build model development stack products, you should be around with customers and the place maybe is Bay.
https://twitter.com/alexgraveley/status/1659276299091812353
This moment has been decades in the making. Six decades of Moore’s Law have given us the compute horsepower to process exaflops of data. Four decades of the internet (accelerated by COVID) have given us trillions of tokens’ worth of training data. Two decades of mobile and cloud computing have given every human a supercomputer in the palm of our hands. In other words, decades of technological progress have accumulated to create the necessary conditions for generative AI to take flight.
ChatGPT became the fastest-growing application with particularly strong product-market fit among students and developers; Midjourney became our collective creative muse and was reported to have reached hundreds of millions of dollars in revenue with a team of just eleven; and Character popularized AI entertainment and companionship and created the consumer “social” application we craved the most—with users spending two hours on average in-app.
Towards Act Two
These applications are different in nature than the first apps out of the gate. They tend to use foundation models as a piece of a more comprehensive solution rather than the entire solution. They introduce new editing interfaces, making the workflows stickier and the outputs better. They are often multi-modal.
The market is already beginning to transition from “Act 1” to “Act 2.” Examples of companies entering “Act 2” include Harvey, which is building custom LLMs for elite law firms; Glean, which is crawling and indexing our workspaces to make Generative AI more relevant at work; and Character and Ava, which are creating digital companions.
This reflects two important thrusts in the market: Generative AI’s evolution from technology hammer to actual use cases and value, and the increasingly multimodal nature of generative AI applications.
The moats are in the customers, not the data. the data that application companies generate does not create an insurmountable moat, and the next generations of foundation models may very well obliterate any data moats that startups generate. Rather, workflows and user networks seem to be creating more durable sources of competitive advantage.
In short, generative AI’s biggest problem is not finding use cases or demand or distribution, it is proving value. What are you going to use all this infrastructure to do? How is it going to change people’s lives?” The path to building enduring businesses will require fixing the retention problem and generating deep enough value for customers that they stick and become daily active users.
If you build model development stack products, you should be around with customers and the place maybe is Bay.
https://twitter.com/alexgraveley/status/1659276299091812353
Sequoia Capital
Generative AI’s Act Two
One year ago, we published a hypothesis that generative AI would become a profound platform shift in technology. Then came the firestorm.
empiricism is the key to progress
rationalism is the key to sounding smart
rationalism is the key to sounding smart