Sonya Huang 🐥
@sonyatweetybird
Followers
15K
Following
4K
Media
54
Statuses
428
New podcast alert!. Join us as we train our neural nets on the theme of the century: AI. Hosted by me, @gradypb, @stephzhan, @kostabuhler and more of our @Sequoia partners. Training Data is a passion project years in the making. The founders and researchers in our orbit are
10
22
142
Introducing the @sequoia Gen AI Market Map!🌎 We’ve decided to map out this emerging frontier, thanks to all the contributions and feedback we’ve received. This space is moving quickly – this map is a living document, so keep the suggestions coming! Who else should we include?
385
1K
7K
Introducing V3 of the @sequoia Generative AI market map! . Generative AI's “Act 1” came from the technology-out. We discovered a new “hammer”—foundation models—and unleashed a wave of novelty apps that were lightweight demonstrations of cool new technology. The market is
61
185
948
As promised: here is our @sequoia Gen AI market map V2! Very thankful for the hundreds of folks who wrote in additions and edits. AI people are amazing 🥹
66
208
858
Platform shifts open the door for founders to build something new and legendary, and we're seeing a Cambrian explosion in applications. Thank you @Bloomberg for having me on to talk about what we're seeing in the AI market at @sequoia!
19
66
364
🥹🥹🥹 It was an absolute joy talking to @bizcarson of @protocol about how we co-wrote our @sequoia Gen AI thesis with GPT-3 (and how I am bad at Twitter, among other things). Thank you, Biz!!. PS: hard at work updating the market map, ready monday :).
13
33
333
We just hosted our 2nd annual @sequoia AI Ascent, a gathering of the 100 top founders and researchers in AI. Big thank you to our amazing speakers including @sama @billpeeb @polynoamial of @OpenAI, @arthurmensch of @MistralAI, @DanielaAmodei of @AnthropicAI, @zoink of @figma,
10
43
333
Congrats team @glean on launching Glean AI! . LLMs are more reasoning engines than knowledge bases. Generative AI is most powerful when tied with an external source of data/knowledge. Glean points Generative AI at your company's enterprise knowledge:.- pre-crawled for speed 🐌
10
30
289
The System 1 vs System 2 thinking framework has really helped me grok what’s happening in the reasoning layer in AI. h/t @polynoamial for first introducing me to this years ago and consistently believing in it and driving it forward. System 1 or “Thinking Fast” = fast
12
48
289
Excited to announce that @sequoia has led the Series A in @LangChainAI 🦜. Some more thoughts on why we invested here:. @hwchase17 is not only insanely humble but also a true visionary on turning LLMs into agents and reasoning applications. Since the early.
12
16
279
1/ Our mission @Sequoia is to help the daring build legendary companies. Now more than ever, we see that opportunity in LatAm.
7
36
267
Episode 1 of our @sequoia AI podcast, Training Data, airs today! 🎬 Featuring @hwchase17 of @LangChainAI w/ my co-host @gradypb in a must-listen episode on agents 🤖. Listen & subscribe (new episodes weekly):. Harrison has been a visionary on turning LLMs.
6
45
212
"BUT WHERE'S THE MARKET MAP?" 🗺️🧐. For our third annual @sequoia generative ai letter, @gradypb and I thought we'd do something a little different. This year, we are *crowdsourcing* our market map, with the simple prompt:. What are the companies that have the best chance of
Introducing V3 of the @sequoia Generative AI market map! . Generative AI's “Act 1” came from the technology-out. We discovered a new “hammer”—foundation models—and unleashed a wave of novelty apps that were lightweight demonstrations of cool new technology. The market is
26
24
209
Once a year, @gradypb and I sit down with our trusty AI collaborators 👾 and zoom out to the big picture on what’s happening in Generative AI. Here’s our 3rd annual take…. 1: The foundation model layer of Generative AI (large, pre-trained language models) is stabilizing around
9
46
191
The latest AI market survey from @retool has some great data. Love seeing @huggingface and @LangChainAI top the AI dev tools charts!
How are companies *𝐚𝐜𝐭𝐮𝐚𝐥𝐥𝐲* using AI? We surveyed 1500+ devs, builders, and business leaders to explore production use cases, the models, infrastructure, and tools people are using, and much more. Dig into the new report: .
2
26
172
Our most exciting episode of Training Data yet 🍓🍰. @OpenAI’s o1 represents a major leap forward by giving models time to "think.". Inference-time compute is the next big research frontier. Thrilled to have @polynoamial, @ilge, and @hunterlightman on the show @gradypb @sequoia
7
26
168
We @Sequoia are excited to double down on our partnership with @FireworksAI_HQ @lqiao and lead their Series B. The AI inference market is inflecting sharply, and Fireworks is leading the market shift to compound AI systems.
5
15
135
🤖⚡️ Thrilled to announce that @sequoia has led @streamlit's Series B, and I'll be representing Sequoia on the board. Excited for a future of beautiful, interactive data apps 🎈. Read more:
6
7
125
Oasis = first fully interactive world model. Instead of being trained words->pixels, the model is trained actions->pixels. This means generative video-like outputs (a la Sora), but you can move around in the world and play it like a video game. And it's real-time. Totally.
We @sequoia are thrilled to announce our investment into @DecartAI. Today they’re launching a “warm-up” called Oasis. This is a fully playable version of Minecraft WITH NO GAME ENGINE, it’s purely a video model rendering pixels in real time
9
19
112
Copywriting tools like @copy_ai @heyjasperai @getanyword @ContendaCo @regieai @copysmith_ai @hypotenuse_labs make it easy to generate blog posts, product descriptions, landing pages, Tweets, and more. Who needs a $200K ghostwriter when you have 👀.
3
5
104
Glean laid the foundation for enterprise search. Now, LLMs are unlocking its value. Here is a wonderful profile of @jainarvind's arc, from star Google search engineer in 2003 to building @glean.
4
2
37
Text-to-image models make us feel like the next Picasso or Studio Ghibli. @OpenAI @midjourney @StabilityAI @WOMBOAI1 @craiyonAI each allow us to create images with a different mood or timbre. Up next: more sophisticated editors, social experiences, search engines, and more.
7
8
98
I'm delighted to share that we @Sequoia are leading @MeetApollo's $110M Series C! Apollo is making intelligent, data-driven GTM accessible for everyone. We were captivated by their product-led growth and raving customer reviews. Congrats Tim, Ray & Apollo!
1
7
86
"Every industry is getting reinvented by A.I.” @sonyatweetybird discusses this watershed moment for technology on @Bloomberg. Watch here:
6
4
86
Hearing @raiza_abubakar describe NotebookLM's 💩💨💩💨💩💨 viral moment was hilarious. Raiza is whip smart and refreshingly real. it's no wonder she's behind @Google's biggest Gen AI success so far. Listen to Raiza & @jayspiel_ on this week's episode of Training Data 🎥
6
6
76
Foundation models have shaken up the digital world. but what about robotics?. On this week’s Training Data, @stephzhan and I spoke to @DrJimFan about @Nvidia's Project GR00T and his quest to build a generalist "foundation agent" that transcends skills, forms, and realities.
5
14
74
Bravo to the @NotionHQ team on creating one of the most elegant and thoughtful Gen ai experiences I’ve seen yet 😍
6
3
75
Last week @FactoryAI announced a new record on the AI coding benchmark SWE-bench. This week, founders @matanSF and @EnoReyes talk about their vision for autonomous Droids, Archimedes' compound lever, and their latest SWE-bench results on our new @sequoia AI podcast, Training Data
3
9
72
When Jim Gao first saw #AlphaGo, he went to @demishassabis at @GoogleDeepMind with a bold proposal: let reinforcement learning run wild on Google's data centers. Demis accepted, and the results were astounding: 40% data center energy savings. One of my favorite episodes of
8
12
71
It’s exciting to think about how AI can help rapidly concept fantastic new worlds in film, television, and more. @SALT_VERSE is creating a 70s style scifi multi-plot film, built entirely with generative models.
2
7
70
Marketers can’t have all the fun – the rest of us get writer’s block too. What about a new text editor with generation at its core? Think Google Docs that can generate entire paragraphs for you. @Get_Writer #Moonbeam #Lex are exploring this space.
3
3
66
60% of @sequoia's seed investments this year have been in AI. 🌲✨. More from @roelofbotha below. The full interview @BloombergTV is worth a watch!
0
8
60
Now that machines can read, they can auto populate companies' knowledge bases with semantic understanding. This means Gen AI can answer questions and file and retrieve documents - a smart, self organizing workspace. @gleanwork @memdotai
3
5
57
Enterprise search is a beast of a problem, but @GleanWork has finally cracked the code on organizing knowledge work. Glean's user love and engagement rivals the best enterprise productivity tools. We @Sequoia are thrilled to lead their $100M Series C.
3
6
57
Physicists are the secret weapon of many great ai research institutions - and we are lucky to have our own AI researcher x physicist in @danintheory. Here is Dan on the physics of AI - worth the read!.
Extreme scenarios of AI focus on what is logically possible rather than what is physically possible. What does physics have to say about AI risk?. 1/2
1
2
21
On our new @sequoia AI podcast Training Data, @LangChainAI founder @hwchase17 talked to us about the idea of a middle ground for AI agents that are in between hard-coded chains and fully-autonomous agents that fly off the rails. Constraints are key to a “goldilocks” zone that are
1
14
57
Design - from interiors/architecture to software to physical products - is up for reinvention. Moodboarding, inspiration, iteration should become drastically faster and better. Less time wasted pushing pixels or organizing artifacts manually. @diagram is a nice example here.
2
2
53
In 2019 Google researcher @fchollet wrote a paper defining intelligence as skill-acquisition efficiency–the ability to learn new skills as humans do. To make it testable he proposed a new benchmark, the Abstraction and Reasoning Corpus (ARC). @Zapier co-founder @mikeknoop read
5
12
53
Video - one of the richest but hardest-to-master mediums of communication - is getting democratized. @runwayml is making it possible for anybody to create Hollywood quality movies with text-to-video, text-based colorization and more.
1
4
50
Gaming experiences may be generated by users or machines on the fly – with AAA level quality. @AiDungeon was one of the first projects to imagine the experience using text-based models, and image and 3D should be ready soon. Here is “Raccoon Heist” from @simonwu:
2
3
45
My favorite part of @hwchase17's Training Data episode is his section on cognitive architectures. Harrison: You can think of cognitive architectures as a "graph" where you've encoded your mental model into the application logic. This graph helps to guide and constrain agents at
2
7
46
I’m also on cloud nine from a week of conversations with incredible AI founders! @gradypb @KostaBuhler @romie_boyd @charliecurnin and I are holding @sequoia AI Office Hours as a way to meet more founders & builders: please sign up here!
13
4
46
1/ Congrats to @Streamlit on joining forces with @SnowflakeDB. Snowflake pioneered the cloud data movement, and Streamlit transformed the ability to visualize and interact with data. We believe this combination will accelerate the field of BI.
3
3
42
Gen AI is unleashing a creativity explosion ⚡️.
1
6
37
New Training Data episode drop! Today we're excited to feature @EricSteinb, one of the great AI researchers of the current/next generation, from the epic story of how he became @polynoamial's collaborator as a high school student to advice for young researchers.
3
8
39
Congratulations @jainarvind and Team @glean! It has been an absolute joy seeing you become the leading enterprise AI assistant, beloved by knowledge workers and AI teams and IT teams alike. It's been a whirlwind year, and the best is yet to come :-).
Five years ago, we set out to build @Glean with the simple goal of helping people find the answers they need to do their best work. Today, we're excited to share that we’ve raised over $200M at a $2.2B valuation led by @kleinerperkins and @lightspeedvp!
2
4
37
The north star for @FireworksAI_HQ is simple API access to the totality of knowledge. The big unlock? Compound AI systems orchestrating 10's/100's/1000's of models and components. On Episode 9 of Training Data from @Sequoia, @gradypb and I spoke to @lqiao about compound AI.
4
3
37
These apps provide an interesting glimpse into what the future may hold. Once you see a machine produce complex functioning code or brilliant images, it’s hard to imagine a future where machines don’t play a fundamental role in how we work and create.
We are entering a new era of human-machine co-creation. If we allow ourselves to dream, then it’s easy to imagine a future where Generative AI is deeply embedded in how we work, create and play. 👽🤠. Excited to share our musings on the space here:.
1
3
34
🤠 @LangChainAI @hwchase17 making moves!! langchain gets you from 0 → 1 with LLMs. and now langsmith takes you from 1 → production with a unified platform to debug, test, evaluate, and monitor your apps. Amazed by this team's ambition and SPEED in the fast-moving ai field.
🦜🛠️ Introducing LangSmith 🦜🔗. A unified platform to help developers debug, test, evaluate, and monitor their LLM applications. Integrates seamlessly with LangChain, but doesn't require it.
1
4
32
Had a blast chatting data apps at #SnowflakeBUILD with #StefanWilliams and @ttunguz. Thank you @SnowflakeDB for having us at the ❄️coolest❄️ event in data!
1
0
31
Congratulations to @tachim and @vladtenev on the Series A @HarmonicMath!. On this week’s episode of Training Data, @gradypb and I spoke to Tudor and Vlad about mathematical super intelligence. We are entering AI's Reasoning Era, and math is the foundation for all reasoning.
4
8
29
Speech Gen allows us to immerse ourselves in experiences that bring out the emotional and expressive qualities of human speech. @play_ht is bringing some of these legends to life in podcast form. We're deep in the uncanny valley with this one.
2
2
29
happy birthday @LangChainAI! seeing you evolve from cool oss library to one of the fastest growing software companies in our portfolio has been an absolute joy :). @hwchase17 @ankush_gola11 @j_schottenstein @romie_boyd.
🎉🎂 LangChain turns 2!. Two years ago, we launched a Python package called `langchain` to help developers build LLM applications that could reason. Today, we're still chasing that same mission — and as the generative AI ecosystem has evolved, we've evolved too. LangChain has
3
2
30
Exciting times for the modern data stack! And now the last mile -- dubbed "operational analytics" -- is here. Welcome @getcensus @borisjabes to the @sequoia fam 💫. The modern data stack: @SnowflakeDB @getdbt @fivetran @LookerData @getcensus.
1
6
26
Most popular agent framework?. LangGraph by @LangChainAI leads by an order of magnitude in production (based on NPM install data), powering agents like @Replit . Major respect to @hwchase17 for shipping not one but TWO market leading frameworks, serving a very different ICP.
Because langgraph is a production ready agent framework… it’s very low level, highly controllable, no hidden prompts. Helps you get to production without getting in the way, so high usage. But a bit of a steeper learning curve (we’re working on that!) so less GitHub stars.
0
2
30
3/ We @Sequoia are actively looking to meet entrepreneurs in LatAm who are building the next generation of enduring companies. If you are a founder in LatAm and would like to meet, shoot us a note at latam@sequoiacap.com. We can't wait to meet you. 🌲.
4
4
27
Biggest “what’s next” for LLM infra: self improving LLM apps. making the current way we build LLM apps look quaint. go @LangChainAI !.
IMO this is the future for a lot of GenAI apps. Automatically achieve better & more personalized LLM apps by:. - Log all LLM outputs.- Log feedback for those outputs.- Use feedback to build datasets.- Use datasets as few shot examples. Want to do this for your app? Shoot me a DM.
2
5
25
Can AI agents outperform the world's best hackers? @XBOW founder @oegerikus thinks so. Oege created @github Copilot and is now building the world's first ✨AI cyber warrior✨ that matches human experts in finding security vulnerabilities. XBOW is one of the first great
2
6
26
It's been a BIG week for AI. Excited to discuss the implications for builders & startups at @axios AI Summit with @AnjneyMidha @imkialikethecar. Tune in tomorrow at 4:45p for the livestream! 🎶.
Catch @sequoia Capital partner, @sonyatweetybird and @a16z General Partner @AnjneyMidha in conversation with @Iamkialikethecar at the #AxiosAISummit in San Francisco. Tune in at 4:45pm PT on Nov. 8:
24
12
24
tune in to our ai ascent playlist on youtube! @hwchase17 on agents is a must-watch.
We've added new videos to our AI Ascent playlist. Hear thoughts on the future of AI from @LangChainAI 's @hwchase17 and @LightmatterCo's @theanalognick. 📺
2
4
26
A treasure trove of lessons from a treasure of a human being. Lucky to be on your team, @gradypb 🍀.
Saturday was my 15 year anniversary @sequoia. one of the things I love about Sequoia is getting to learn new things from amazing people, so here are 15 lessons from 15 people over 15 years. hope you enjoy!.
2
1
25
Congrats @census @borisjabes !!! The modern data stack @snowflakeDB @getdbt has forever changed the way we store and analyze data. The next frontier is taking action on that data, in all of your tools. Excited for the role @census is playing to paint that future.
So excited to announce our $60M Series B! It's been a wild journey but every year gets more interesting, especially when growing 10x :-). I love the work we do @census – w/ operational analytics we make data teams the linchpin of their organization!.
1
7
23
Last week @DecartAI introduced Oasis, the world's first interactive real-time video model – a video game entirely generated by AI – and went instantly viral. But @DLeitersdorf says Oasis is just a baby step toward a new wave of experiences, including a magic mirror.
1
5
25
Excited to see AI projects graduating from blue-sky ideas & weekend hackathons into production. @fintual is tackling one of the most interesting (and meaningful) conversational experiences out there: a user-facing financial advisor built on LLMs. [and built and hardened with.
Fintual Copiloto.La primera solución de inversiones personales con inteligencia artificial en México. 3 de agosto. Daremos acceso primero a las personas que se inscriban en
6
6
23
Congratulations @nubank on $NU! We’re so honored to have been part of the journey with @velez_david from day one to #NuDay.
1
2
22
I've been waiting for AI to disrupt the traditional BI market with a "talk to your data" product for a long time. @RamaswmySridhar told us about how @SnowflakeDB is using system design and engineering to bring "talk-to-your-data" LLM accuracy from 45% (out of the box) to 90%.
2
6
24
So excited to be working with @gastonirigoyen !!! . Fintech is one of the most interesting opportunities in LatAm -- and the infrastructure supporting this boom is antiquated. @pomelo_latam is fixing that. Vamos!!! 🇧🇷🍊.
Sequoia invierte en Pomelo y desembarcamos en Brasil 🚀🇧🇷. Hoy anunciamos U$D 1 millón de inversion adicional, elevando nuestra ronda seed a un total de U$D 10 millones.
2
1
23
Woo hoo go @streamlit !!!.
Several startups on the enterprise AI list are trying to make it easier for data science teams to work with machine-learning models, including @huggingface, @appzen, @OctoML, @focalsystems, @hypersonix_ai and @streamlit 🔍 #ti50.
1
1
23
We are entering AI's reasoning era. and @FireworksAI_HQ just did a mic drop by dropping the best reasoning model. Fireworks f1 is the first reasoning system over open models to beat GPT-4o and Claude 3.5 Sonnet across hard coding, chat and math benchmarks. How did they do it?.
🔥 Introducing Fireworks f1 🔥 . f1 is the first reasoning system over open models to beat GPT-4o and Claude 3.5 Sonnet across hard coding, chat and math benchmarks. 💥 two variants now available in preview: f1 and f1-mini .💥 access the preview on Fireworks AI Playground for
2
0
23
In the latest episode of Training Data (@sequoia's AI podcast), @Microsoft CTO @kevin_scott reflected on the early days of the @OpenAI partnership and their shared conviction in scaling laws. Tune in to hear more about the beginnings of their epic partnership.
2
2
23
Congratulations @jainarvind and @glean! It's been a joy to see you evolve from ambitious technologist to CEO of one of the best enterprise companies of the AI era. Can't wait to see what's ahead for Glean -- first search, then assistant, then apps. agents next? :-).
Today, I’m excited to share that @Glean has raised over $260M in new funding at a $4.6B valuation co-led by Altimeter and DST Global, as well as launched exciting new product capabilities that together will enable Glean’s Work AI platform to expand its reach and deliver everyday
2
1
23
LLMs are trained on internet-scale data. But to make AI work for work, models need to be grounded in your enterprise data. @glean spent 5 years building the ingestion, search, and ranking infra needed to solve this problem well. CEO @jainarvind shares more on Training Data.
3
5
23
On our latest episode of Training Data, @EnoReyes of @FactoryAI shared this pearl of advice for AI founders: . "Build something that you think is going to get 10x better when OpenAI releases GPT-6 or 7. We think of our product as something that will multiply in value and
1
0
21
How will society navigate AI's impacts on employment? @klarna already has AI performing the work of 700 customer service reps, with higher CSAT. But what about the human impacts?. @klarnaseb shared a nuanced take on the path ahead on episode 6 of @sequoia's Training Data.
2
3
20
On the latest episode of Training Data from @Sequoia, @stephzhan and I spoke with @GitHub CEO @ashthomas to hear how a small team at GitHub built on top of GPT-3 and quickly created a product that developers love—and can’t live without.
1
2
20
Taking long context to a whole new order of magnitude. Congratulations on 100 million @EricSteinb! and excited to finally make things official :).
LTM-2-Mini is our first model with a 100 million token context window. That’s 10 million lines of code, or 750 novels. Full blog: Evals, efficiency, and more ↓.
0
0
19
a powerful read & a reminder to be kind 💜 -- you never know the hidden challenges those around you are living through. @nataliemiyake you are tough, and you are going to be an AMAZING mother!.
After four years on the IVF journey, we’re thrilled to share that we are finally expecting a baby girl in February via surrogacy. While no two journeys are the same, I want to share what @KyleB and I endured to arrive where we are today. 1/.
0
1
18
HOORAY @streamlit custom theming & dark mode is here!!! excited to unleash my inner picasso 👨🎨🎨🌘.
Introducing Custom Theming 🌈 and Dark Mode🌛. Upgrade to easily toggle between Light Mode, Dark Mode or your own Custom Theme!. 📖 Read more: 🧪 Sample app: #datascience #python #theme
1
1
17
Ever since @klarnaseb tweeted about how @klarna got AI to handle the work of 700 customer service reps, every company is asking “How do we get the Klarna customer support thing?”. He joins us on this week’s Training Data to explain how they did it. LINK:
4
3
15