Andrew Hon

🚀 We've open-sourced TimesFM – a decoder-only foundation model for time-series forecasting 📈 Key Highlights: ✅ Large-Scale Training Data: Trained on a corpus of over 100 billion real-world time points, including Google Trends and Wikipedia pageviews. ✅ Zero-Shot Forecasting: Offers high-quality forecasts even on unseen datasets, without requiring fine-tuning or additional training. ✅ Outstanding Benchmark Performance: Achieved competitive results across a comprehensive suite of benchmarks. 🤗 Explore the model on Hugging Face and GitHub. I'm encouraged by the early feedback and look forward to seeing how #TimesFM will be used in real-world applications for forecasting and anomaly detection. 🙏 A special thanks to the talented research team behind this achievement: Abhimanyu Das Rajat Sen Yichen Zhou Weihao Kong

42

I am so proud of the work that is being delivered by our team in Search that was showcased at I/O this week. Highlights were... ⚡Plan Ahead with Generative AI in Search ⚡ Thanks to a new Gemini model customized for Google Search, we are able to do SO MUCH more. With planning capabilities *directly* in Search, you can get help creating plans for whatever you need, starting with meals and holidays. Search for something like “create a 3 day meal plan for a group that’s easy to prepare,” and you’ll get a starting point with a wide range of recipes from across the web. And if you don’t like the suggestions, Search will continue working on your behalf. So if you want to swap the dinner suggestion for a vegetarian option, just like that, Search will customize your meal plan. On top of this, there are lots more updates in Search so — for the curious  — have a read of the blog by Head of Search Elizabeth Reid below 🔽 https://lnkd.in/en6_7fvX #GoogleIO #GenAI #GoogleSearch

177

7 Comments

Common Room's latest launch—RoomieAI™—is a big one. It’s not just one thing, it’s a suite of AI-powered tools built with modern go-to-market teams in mind. ✨ Ask RoomieAI Our AI agent will help you deep dive into account research—super fast and without the tab-hopping reps have to put up with today. Need to run some competitive analysis? RoomieAI will explain a target account’s market positioning and serve up their top competitors. Looking to extract insights from recent financial filings? RoomieAI will summarize the top takeaways from a 10-K so you can craft better outbound. All the intel you want is just a prompt away. Use our out-of-the-box options or create your own. ✍️ RoomieAI message snippets In private beta now (get in touch to request early access!), RoomieAI message snippets will help you go from blank page to impactful outreach lightning-fast. No more wasting time trying to piece together the right signals and turn them into personalized, relevant messages. RoomieAI can do the heavy lifting for you by turning the buyer actions and attributes captured in Common Room into ready-to-send outbound snippets. Add them to your sequencer tool and nail the right message right away. There are lots of reasons I’m excited about these new features, but the biggest one is this: It helps our customers tap into their ultimate differentiator, customer intelligence. Lots of tech vendors are racing to scale outbound using the same datasets every broker offers. But what’s the point of automating the same stale info? Using AI to personalize attributes like company size, role, etc., might be better than randomly dropping someone into a prospecting sequence, but it’s not going to help you stand out. Especially not if everyone else is doing the same thing. Common Room fixes that problem. How? Comprehensive signal capture across 1st-, 2nd-, and 3rd-party data sources and our proprietary identity resolution and enrichment engine, Person360™. When you capture more signals—and connect them to real people and real accounts—you have more and better data to train an AI model with. That means our new generative AI tools can build messaging with more context, more personalization, and more relevance. Proprietary data and the right set of signals is how you create truly differentiated outbound. But I’ll step down from my soapbox 😉 Check out the link in the comments for more info ⬇️

38

2 Comments

Thanks to Phuong Pham for pointing me at this helpful, relatively recent data from OpenView and ScaleXP. The graph here shows median (22 months for companies $50M and up) and top quartile (just 14 months for that same set) CAC payback periods. An article worth bookmarking for later reference, my growth stage Sales and Marketing friends. https://lnkd.in/gm8s7TgA

16

2 Comments

My bets where AI won't replace humans: 1) Human connection: Like the one that many of us just had at Config. 2) Shared human experience: When you look at another person's eye and you know you've been through a similar joyful and painful journey. 3) Real stories behind art and culture: I listen to music, and I like paintings not just because they feel good, but because there's a real human story behind it that speaks back to 2. 4) Ingenious net new creative fields: AI can use past things to create something new by mixing them, but it's mostly always can't really create something completely new and unseen. 5) What's your bet?

34

4 Comments

Excellent study from Damian Rollison and the SOCi, Inc. team which seems to confirm my longtime hypothesis around citations (https://lnkd.in/gxZQwdx3): the only citations that are valuable are the ones that rank in Google for the keywords you want to be known for. https://lnkd.in/g49DMSeN

39

11 Comments

To all my British friends, this intro looks like the #LoveIsland Villa. But disappointingly, I only speak about the very unsexy topics of data hygiene, values, and the nitty gritty of how your team preps to use your company's data as the foundation of AI. If you want to talk about how data governance as team effort, DM me or just heckle me in the comments. "I've got a text!"

9

1 Comment

Unsurprising to anyone that knows me well, I love to perform analysis of product efforts, sure platforms like GA, mixpanel, adobe, etc are fun to quickly capture trends, but the joy comes from writing out the db queries and getting deeper into the signals. Getting deeper into the numbers as a start-up is a double-edged sword: I CAN join this table to add another column, I CAN look at more numbers, I CAN find more signals. However, in a retrospective yesterday we had a great discussion on diminishing returns within noise to signal. In the first pass, is there a trustworthy signal that can be used for decisions? If yes, then share the report and the signal and move on to the decision! More numbers could very well just be noise, or it could simply be an hour of work that just added a 1% support of the original signal. And you don't have enough time at a startup to build the huge report, tease out the noise, decide, defend the decision signal that you found through the noise, build the decision, and build the other unrelated necessities on the roadmap. Find the signal, validate the signal, decide, and move on. Did I write a 500 line query? yes! Does it provide a strong signal? yes? Did a 50 line query reinforce and provide the same signal but brighter? sure did. Do less so you can do more

16

1 Comment

Our team just release a new benchmark for Hand and Object Tracking called Hot3D: https://lnkd.in/gBT2bF5H. HOT3D is a new dataset to explore vision-based methods for hand-object interaction. And we believe it will unlock new opportunities within this research area, such as transferring manual skills from experts to less experienced users or robots, helping an AI assistant to understand user's actions, or enabling new input capabilities for AR/VR users, such as turning any physical surface to a virtual keyboard or any pencil to a multi-functional magic wand. To learn more about the dataset, read the HOT3D paper (https://lnkd.in/gqp34kZa) and visit https://lnkd.in/gUy4RUYX. HOT3D will be part of the BOP challenge (https://bop.felk.cvut.cz/) and the HANDS challenge (https://lnkd.in/gQnbSBih) at ECCV-2024. Check out the respective websites for more details on the challenges. #CVPR2024 #ComputerVision #AI

788

13 Comments

I'm finally ready to share an experiment I've been working on with the Future Audiences team at Wikimedia. It's a pretty special team where I get to have space to think about the intersection of things like LLMs and the evolving online space of (mis/dis)information: specifically how can this technology not replace humans as an information source, but help better connect us to living sources of information created by humans. I have this hope/dream that tools for us to verify information in a world where creating questionable information is so widespread will be as easy and normal as your internet browser coming with built-in adblockers to protect you from being overwhelmed with pop-up window spam. Anyway, the experiment we have is called Citation Needed, and if you are using the Chrome browser, I'd love for you to try it out. After you install it, you can use it to check short texts online to see if they are corroborated by information on Wikipedia, and have the opportunity to learn more about it. It's still experimental, and uses LLMs/AI, so you might see some funny behavior once in a while -- which you can report to us in the extension -- though I invite you to imagine what this functionality could look like in the near future, and/or beyond the confines of a browser extension. And then let us know where this sci fi daydream takes you. Link: https://lnkd.in/eQwd4iqK

96

8 Comments

🚀🚀 Exciting news! 🚀 🚀 I left Google two months ago to start a new company - Aligned Labs - with Tyler Garvin At Google, I had the unique opportunity to work at the intersection of product, design, and strategy for our large language models, like Gemini. This experience made one thing crystal clear: data makes the difference. As we push the boundaries of AI capabilities, we're rapidly depleting our sources of unique, high-quality data. This isn't just a speed bump—it's a potential roadblock for the entire industry. At Aligned, we're not just collecting data—we're curating a first-of-its-kind team of the world’s leading domain experts and educators. Our mission is to create top-tier, unique datasets for training large language models. Our goal? To ensure AI outputs align with our expectations of highly intelligent thought partners. We're crafting datasets for every stage: pre-training, instruction-tuning, fine-tuning, and evaluation to empower our clients to unlock the full potential of their models and applications. I'm dedicating the rest of my career to this for three critical reasons: - The data crisis is real and imminent. Without a solution, LLM performance will plateau, stalling AI progress. - As models advance, they require increasingly specialized data. We need to tap into expert knowledge across various domains, but no one has cracked the code on how to scalably attract and retain this talent. Companies like Scale AI have done an amazing job of building a workforce of cheap educated workers but what you really need for the next level of performance are domain experts that can train and impart their expertise and knowledge into these models. - Frontier model builders like Google, OpenAI, Meta, and Anthropic understand this problem and are making massive investments, there's a huge opportunity in specialized data. That opportunity goes beyond the Google’s and OpenAI’s of the world. High-quality, unique datasets can dramatically improve AI applications in specific industries, making them perform better and more cost-effectively. We're here to unlock this potential, helping companies of all sizes harness the power of tailored, premium data to build superior models and applications for their specific needs. We're incredibly excited about the potential of our work to dramatically improve AI models and help companies across industries harness the power of truly intelligent AI. Are you passionate about the future of AI? Let's connect! Whether you're a potential client, an AI enthusiast, or a domain expert interested in contributing to the next wave of AI advancement, I'd love to hear from you. Head over to https://www.alignedhq.ai to connect! #AIStartup #MachineLearning #DataQuality #FutureOfAI #AlignedLabs

280

75 Comments

10 months ago, I embarked on this journey with a simple yet ambitious goal: to create an AI assistant that knows you like your best friend and can truly enhance your life. Now, Ario is in open beta! 🎉 This journey has been filled with - Joy from hearing beta users say, "Honestly, I'm stunned, like wow," - Excitement from rounds of rapid iterations - Sudden enlightenment in solving unexpected challenges while building our AI agent Every step has brought us closer to creating something truly special - Ario's Personalized, Proactive and Private. I’d love for you to give Ario a try and let me know your thoughts. Your feedback helps us shape Ario. And if you're also navigating the fascinating world of AI, let’s connect and exchange insights!

20

1 Comment

PSA: Stop using the original OpenAI CLIP weights. I see a lot of people using CLIP for multi-modal retrieval. What many people may not know is that there are 100’s of CLIP models now since OpenAI’s original release. These newer models offer significantly better performance across many benchmarks. For product search related tasks we recommend the following. + Fastest inference: ViT-B-32 [laion2b_s34b_b79k] + Best balanced: ViT-L-14 [laion2b_s32b_b82k] + Best-all round: xlm-roberta-large-ViT-H-14 [frozen_laion5b_s13b_b90k] You might notice these are not the ones that perform the best on the benchmarks. Why? Because the benchmark only test for very specific tasks which are often quite different from search and recommendations. These are ones we have found to be the best for product related search and retrieval tasks and generally very good first choices if you are unsure where to start (even for non-product related search). Here is a hugging face space with all the data: https://lnkd.in/gxu5UYkt Read more: https://lnkd.in/g5_D4ccs

293

22 Comments

Isar Meitis has does a great job creating content covering not only the what of AI use cases but spending time diving into the how. I had a great time chatting with him for this episode that went into details on one of the most entry-level AI + Automation use cases I share and then into some of the most advanced ones. We also got a chance to talk about Zapier's first agentic tool, Central (Which you can try for free today!) All the links in the comments here. 👇

21

2 Comments

Check out this article written by our amazingly talented data science team on how we developed a statistical engine for Skyscanner's in-house experimentation platform!

8

📈CLIP benchmarks for multimodal search and recommendations 📊We benchmarked over one hundred CLIP models to assess them not only on text to image retrieval (a proxy for search relevancy) but also: - latency when embedding text or images. - text context length. - image input size. - inference across two GPU’s, the T4 and A10g. We also provide recommendations on which models to use and how to think about trade-offs. Learn more: Blog post - https://lnkd.in/g5_D4ccs HuggingFace space - https://lnkd.in/gxu5UYkt Marqo - https://lnkd.in/gsFRwZT5

33

Last week, in the 100+ years’ old Palace of Fine Arts, gorgeous in the early summer California sun, over 30,000 people and some robots showed up for the GenAI summit https://genaisummit.ai/#/. The somewhat disorganized conference that attracted attendees from all over the world is a testimony of the current state of AI: everyone wants to get in, a LOT is happening, no one really has the paved path, but there is no doubt the future is here. As a student of AI, my key takeaways: ▶ Development of base models will be done by a few players who have the money to burn. One presenter quoted up to $100M for a single model. Though the architecture of the base models may be similar, each player has their secret sauce. And the winner of one model is unlikely to take all. All players are rapidly developing and revolutionizing. 🔹 Model “aggregators” fit domain-specific data to base models. They are not “picky” about which model to use. They add value by fine-tuning the models for specific domains. From prompt engineering to RAG to fine-tuning, SMEs play a key role in “landing” LLMs well. There is a boom of startups in this space. ⬆ Data is still foundational. It’s the era of unstructured data. New data storage and processing technologies are the backbone. Is there “enough” data for AI? It depends. Synthetic data is widely used. While “connected” data on the web is relatively easy for the models to access, “protected” data in enterprises requires jumping many hoops, including expert domain knowledge, governance and “imagining” the business problems they can solve. Data about the real world is fast growing, collected from sensors and fed into robotics. 💲 GPUs are expensive and power hungry. Optimization is the path today. Alternative hardware can take time. The bigness of AI meets the smallness of nuclear power. Mini nuclear power plant https://lnkd.in/giJkjTiG. 🛠 While the base model developers are relatively a small elite group, often with deep academia roots, applying the models to real world problems needs everyone in their current job working with AI. AI certainly created new job titles, but also makes it increasingly easier for the public to use it as a new “tool”, instead of a complete career change. 🐬 The most optimistic prediction of AGI is 3 years. The honest answer is no one really knows. How far will AI go from data, information, knowledge, intelligence to wisdom? Opinions differ. When the answer is 42, how would we know 😉 ? 

16

3 Comments

If your internet has been broken for the past 24 hours, you may have missed that Triple Whale 🐳🐳🐳 launched another winning product. The product is called Sonar. Here's what Sonar does: 1. It takes all the user journey and site activity data that TW collects 2. And sends it back to platforms like Meta and Klaviyo 3. Enabling their targeting algorithms to operate based on more complete data Better targeting = more orders = better business It's as simple as that. In studies we found that Sonar results in 70% more abandon cart flows being sent by Klaviyo. That's nuts! And the crazy thing is that there's ~no work required to unlock this efficiency hack. You just turn it on, and it works 😍 Want to give it a try? Reach out to your customer success manager, or if you're new to the whale, visit us here: https://lnkd.in/db72z8aA

57

2 Comments

Thinking about how to leverage AI? Unsure which LLM is right, and secure enough, for your business use case? If you are looking to drive CRM results, this will certainly help you narrow the LLM field, with more innovation and enhancements to follow! This benchmark measures the performance of LLMs against four key measures: accuracy, cost, speed, and trust and safety

12