Michael Lai

This San Franciscan has landed in the Big Apple for #NYTechWeek! Here’s why it was important to me to be out here this week and I view NYC as a sister city. ((scroll down for 2 events The Council is hosting today!)) 🏭🏥 Reminder: The Council Fund invests at Pre-seed in Software’s Third Act, the transformation of the essential economy. Our investments span healthcare and industrial sectors like manufacturing, logistics, construction. 🐂 I’m eternally bullish on NY & SF and hate how many pit the two cities against each other. The reality is that a lot of capital and talent flow back and forth between the cities. They need each other to thrive! 🗽🌁 Cities like NY & SF attract young and mid-career talent because of their population density, transient nature, beauty, and reputations as cities where someone new on the scene can make a name for themselves. For the areas we invest in, I’m *specifically* excited about NY because of: 🏗️ Closer proximity to industrial hubs which are scattered across the Midwest / Rust Belt 🏛️ Closer proximity to DC where a lot of regulation around VC, startups, healthcare, onshoring initiatives, and industrial sectors takes place 💰 Many of the large financial institutions that are backing hard tech focused funds and major healthcare initiatives are also headquartered here in NYC 🎯 We’ve got several startups here who are hitting it out of the park, like Dirac and Siena. This has further validated my belief in this ecosystem to support young companies! Today we are kicking off with 2 events on Day 1 of #NYTechWeek 👉 AI Demo Day: The Council, Next Wave NYC, Lynx Collective, Flybridge will host 8 founders and 200+ investors and founders for a pitch competition this evening! Luma: https://lnkd.in/gbdYiJMw 🔜 Can still join the waitlist! Small & Big VCs in Hard Tech: The Council joined General Catalyst, Lux Capital, Iron Prairie Ventures, AlleyCorp, House Of Ventures, Steel Atlas for 2 back to back panels on industrial innovation & collaboration across the capital stack Luma: https://lnkd.in/gj6D4-Au ⏰ Event has passed

126

6 Comments

One new interesting application of LLMs: CAIS has released a forecasting bot that provides probability estimates for any question you ask. For example, to the question "Will GPT5 be released before 2025?", it answers that it's 70% likely. The bot works by searching for relevant information online, feeding it to GPT-4o, asking a few guiding questions, and then requesting a probability estimate. What's impressive is that this bot outperforms experienced human forecasters, even though GPT-4o wasn't specifically trained for this task. This kind of application shows that AI could be of great help in improving decision-making in the future. Link to the blog post: https://lnkd.in/gjP7JQgz Link the bot if you want to try it yourself: https://forecast.safe.ai/

11

Introducing an AI-generated CIM review for PE professionals in <5 mins: The first time I worked on a Quick Look, it took me a week to create the first deliverable (don’t judge!). Not to mention the editing + iterations that came afterwards. But by their nature, *Quick* Looks are fairly standardized. In this early demo, we’ve transformed a 50-page CIM into the first 8 pages of a PPT deliverable + a fully functioning LBO model. You have a first draft containing content on: -         the Company Overview -         Segments Overview (stacked bar charts + revenue bridge) -         Investment Thesis + Risks (this obviously needs to be refined manually) -         LBO one-pager -         Returns sensitivities -         Management bios -         Appendix: Additional pie charts formatted properly -         + the full LBO model in Excel, customized to the source financials While it’s not 100% perfect quality, I think of this as asking an intern to take the first pass. I know a lot of professionals would appreciate that v1. And it’s not just about populating a simple template: Here, the LLM made important decisions to handle edge cases in the source data. For example, we’ve grayed out certain financials that weren’t provided in the CIM, and adjusted the LBO model to extrapolate accordingly. That way, you get a fully functioning LBO regardless of what the source data contains. If this is interesting to you, please reach out! We have a lot more to come soon -- so stay tuned.

127

4 Comments

🚀 Unlocking the Power of Variational Autoencoders in Time Series Analysis! 📊 I'm excited to share my latest Substack post, where I dive deep into the fascinating world of Variational Autoencoders (VAEs) and their application in generating synthetic time series data. 🔍 What You'll Discover: A clear explanation of how VAEs work, including the encoder and decoder architecture. Step-by-step code to implement a VAE for time series data generation, complete with visualizations of the results! This post is perfect for data scientists, machine learning enthusiasts, or anyone curious about generative models and their practical applications. Let’s unlock new insights in data generation together! 🌟 #MachineLearning #DataScience #VariationalAutoencoder #TimeSeries #AI #DeepLearning #DataGeneration

2

Something I've heard from a lot of customers building Large Language Model applications (especially those building with retrieval augmented generation). The biggest lever for improvement often doesn't come from improving the model or prompt. Its actually comes from diligently fixing the underlying retrieved data chunks. For many knowledge corpuses, there are often many inconsistencies that exist (for example, if a document has an outdated customer policy). This is especially true the larger a corpus becomes - it's really hard to maintain an accurate source of truth across thousands of documents. Improvement comes from rigorous testing and identification of when the underlying retrieved chunk actually contains incorrect factual information - and taking time to align on a single source of truth. This can be automated to some extent, but also requires manual intervention from subject matter experts who actually know the truth (for example, a very knowledgeable customer servicing agent). Something to think on - let me know if you're facing something similar in your work!

31

3 Comments

The company that Cody Coleman, William A. Gaviria Rojas, and the amazing team at Coactive AI are building is incredibly important. They are quickly solving one of the BIGGEST and FASTEST growing problems in data infrastructure today: Unstructured Visual Data - What is it, What to do with it, and How to monetize it. Congrats on the $30M series B!!! Pumped to work alongside Martin Casado at Andreessen Horowitz + Stacy Brown-Philpot at Cherryrock Capital + Brentt Baltimore at Greycroft + Fern Mandelbaum at Emerson Collective in this journey! Coactive AI's Multimodal Application Platform offering is also a testament to how having a diverse team can help customers address a very complex and diverse set of technological & business challenges. More from team Coactive AI below and in the great coverage from my friend Kenrick Cai at Forbes! Read Funding Coverage Here: https://rb.gy/42rsc5 Read more about the Coactive AI MAP Product Release: https://lnkd.in/ed4Gykkz "The world we live in is visual—it’s in every aspect of our lives, from the way we work and communicate to how we shop and enjoy entertainment. Yet, understanding visual data has been elusive, expensive, and tedious for many businesses for far too long. It’s been previously unachievable to intelligently search and analyze large volumes of unstructured visual content without spending countless hours labeling, tagging, or adding metadata to images and video, resulting in lost time and productivity. We are solving this problem at scale. Our mission is to transform how companies understand and analyze their visual content, turning data from a tax to an asset."

168

4 Comments

Any New Yorkers / Brooklynites out there? I'm hosting a free in-person workshop and meetup on October 22nd (week after next) in Brooklyn, NYC! Community has always been a driving motivation behind the Analytics Accelerator. After working individually with almost 100 students over the last year, I've noticed that people tend to get "stuck" when they focus too much on isolated self-learning, without understanding how their work relates to a team context - and how to represent this in the job hunt. Being part of a community not only expands your learning, but builds up your sense of confidence and self-conviction that you can make it to the other side - no matter how competitive the job market. So come be a part of our growing community of ambitious analysts! In this event, I'll share insights on: 1. Most common pitfalls of aspiring analysts (and how to overcome them) 2. How to get creative about getting real data experience 3. How to demonstrate the traits of a standout analyst 4. Examples of highly-successful (and unsuccessful) resumes and projects At the end, we'll open it up to drinks / networking / open questions and discussions about working in data. Join us for a fun and insightful time. RSVP HERE - excited to meet you! (Feel free to reshare if it'd be helpful to your community :) )

24

2 Comments

Great discussion today here at Antler NYC residency on corrective RAG (from this paper: https://lnkd.in/etf-H5Yt The problem: You ask a question to a large language model, it pulls in relevant information to help answer better, but the information that it references is wrong, so its response is wrong as well. The solution: Add a specialized layer that verifies the correctness of the additional context. If the context is valid, the model uses it to help inform its answer. If the context check fails, the model searches the internet and/or additional documents to find the right context before answering. It's not a foolproof system, but it is a step in the right direction to mitigate hallucinations and improve the quality of the LLM product

22

7 Comments

OK, this is the fourth of a short series of posts about forecasting and metrics. Some of what I see "cheeses me off," so this may get brutal and honest. 4. 85TH PERCENTILE DOESN'T MEAN ANYTHING What cheeses me off: Tools and training that says use 85th Percentile for Monte Carlo forecasts. It has no special mathematical or logical reasoning. What I wish you would do: Monte Carlo forecasting lets you opt in to how optimistic or pessimistic you want your forecast. Linear forecasting using averages gives results that are likely 50/50. And that's OK unless the cost of being wrong is catastrophic financially or commercially. I want you to consider how important the date you are about to give is. If not so much, go 50%; if massively impactful, go higher, 95%, or even 100%. BUT THINK ABOUT IT. Why you do this: Forecasting is communication. Its about collaborating with peers on what could happen. Giving a fixed date without discussing the impact of missing that date is an error. Don't use tools that just give dates without discussion. (Backstory of why I think it became folklore: In an Excel spreadsheet, I needed to color code the probabilities; I used pink for 50% and below, then wanted to split the 50-75-100 in yellow and green. I made the greens smaller at the last minute, 50-80 yellow and 85-100 green. 85% was the first green line. If you google it, 85% is used in traffic management. I've been told it's where the elbow is on some distribution curves, but that's all incorrect)

113

25 Comments

How 𝘾𝙖𝙪𝙨𝙖𝙡 𝘼𝙄 Can Help Reduce Hallucinations: New Q&A with Jayeeta Putatunda I recently spoke with Jayeeta, Lead Data Scientist and Director at Fitch Group Inc, about using causal AI to ground language models in real relationships, not just correlations: 𝘛𝘩𝘪𝘯𝘬 𝘰𝘧 𝘧𝘪𝘯𝘢𝘯𝘤𝘦 𝘰𝘳 𝘢𝘯𝘺 𝘥𝘰𝘮𝘢𝘪𝘯 𝘸𝘩𝘦𝘳𝘦 𝘺𝘰𝘶 𝘤𝘢𝘳𝘦 𝘢𝘣𝘰𝘶𝘵 𝘤𝘢𝘶𝘴𝘦 𝘢𝘯𝘥 𝘦𝘧𝘧𝘦𝘤𝘵 𝘳𝘢𝘵𝘩𝘦𝘳 𝘵𝘩𝘢𝘯 𝘫𝘶𝘴𝘵 𝘤𝘰𝘳𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯. 𝘌𝘤𝘰𝘯𝘰𝘮𝘦𝘵𝘳𝘪𝘤𝘴 𝘩𝘢𝘴 𝘢𝘭𝘸𝘢𝘺𝘴 𝘥𝘰𝘯𝘦 𝘵𝘩𝘢𝘵, 𝘵𝘳𝘺𝘪𝘯𝘨 𝘵𝘰 𝘴𝘦𝘦 𝘪𝘧 𝘢 5% 𝘤𝘩𝘢𝘯𝘨𝘦 𝘪𝘯 𝘱𝘢𝘳𝘢𝘮𝘦𝘵𝘦𝘳 𝘈 𝘭𝘦𝘢𝘥𝘴 𝘵𝘰 𝘴𝘰𝘮𝘦 𝘴𝘩𝘪𝘧𝘵 𝘪𝘯 𝘱𝘢𝘳𝘢𝘮𝘦𝘵𝘦𝘳𝘴 𝘉 𝘰𝘳 𝘊. 𝘠𝘰𝘶 𝘤𝘢𝘯 𝘳𝘶𝘯 𝘴𝘪𝘮𝘶𝘭𝘢𝘵𝘪𝘰𝘯𝘴 𝘵𝘰 𝘴𝘦𝘦 𝘪𝘧 𝘵𝘩𝘢𝘵 𝘭𝘪𝘯𝘬 𝘩𝘰𝘭𝘥𝘴, 𝘰𝘳 𝘪𝘧 𝘵𝘩𝘦𝘳𝘦’𝘴 𝘫𝘶𝘴𝘵 𝘢 𝘭𝘰𝘰𝘴𝘦 𝘤𝘰𝘳𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯. 𝘕𝘰𝘸 𝘤𝘰𝘮𝘣𝘪𝘯𝘦 𝘵𝘩𝘢𝘵 𝘸𝘪𝘵𝘩 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘈𝘐. 𝘐𝘧 𝘺𝘰𝘶 𝘧𝘦𝘦𝘥 𝘵𝘩𝘦 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘮𝘰𝘥𝘦𝘭 𝘷𝘢𝘭𝘪𝘥𝘢𝘵𝘦𝘥 𝘤𝘢𝘶𝘴𝘢𝘭 𝘮𝘢𝘱𝘴 𝘰𝘳 𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯𝘴𝘩𝘪𝘱𝘴, 𝘵𝘩𝘦𝘯 𝘸𝘩𝘦𝘯 𝘪𝘵 𝘱𝘳𝘰𝘥𝘶𝘤𝘦𝘴 𝘢 𝘳𝘦𝘱𝘰𝘳𝘵, 𝘪𝘵’𝘴 𝘭𝘦𝘴𝘴 𝘭𝘪𝘬𝘦𝘭𝘺 𝘵𝘰 𝘪𝘯𝘷𝘦𝘯𝘵 𝘧𝘢𝘭𝘴𝘦 𝘤𝘰𝘯𝘯𝘦𝘤𝘵𝘪𝘰𝘯𝘴. 𝘐𝘧 𝘺𝘰𝘶’𝘳𝘦 𝘴𝘶𝘮𝘮𝘢𝘳𝘪𝘻𝘪𝘯𝘨 𝘩𝘰𝘸 𝘈𝘱𝘱𝘭𝘦’𝘴 𝘨𝘳𝘦𝘦𝘯 𝘪𝘯𝘪𝘵𝘪𝘢𝘵𝘪𝘷𝘦𝘴 𝘭𝘦𝘥 𝘵𝘰 𝘢 𝘳𝘦𝘥𝘶𝘤𝘵𝘪𝘰𝘯 𝘪𝘯 𝘤𝘢𝘳𝘣𝘰𝘯 𝘦𝘮𝘪𝘴𝘴𝘪𝘰𝘯𝘴, 𝘺𝘰𝘶 𝘤𝘢𝘯 𝘷𝘢𝘭𝘪𝘥𝘢𝘵𝘦 𝘵𝘩𝘢𝘵 5% 𝘰𝘳 10% 𝘧𝘪𝘨𝘶𝘳𝘦 𝘵𝘩𝘳𝘰𝘶𝘨𝘩 𝘺𝘰𝘶𝘳 𝘤𝘢𝘶𝘴𝘢𝘭 𝘮𝘰𝘥𝘦𝘭. 𝘛𝘩𝘦 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘮𝘰𝘥𝘦𝘭 𝘤𝘢𝘯 𝘵𝘩𝘦𝘯 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘦 𝘵𝘦𝘹𝘵 𝘵𝘩𝘢𝘵 𝘳𝘦𝘧𝘭𝘦𝘤𝘵𝘴 𝘵𝘩𝘰𝘴𝘦 𝘳𝘦𝘭𝘢𝘵𝘪𝘰𝘯𝘴𝘩𝘪𝘱𝘴, 𝘳𝘢𝘵𝘩𝘦𝘳 𝘵𝘩𝘢𝘯 𝘨𝘶𝘦𝘴𝘴𝘪𝘯𝘨. Jayeeta also shares her path from econometrics to NLP innovations, her perspective on the rapid evolution of AI technologies, and the importance of bridging research with real-world industry challenges. Other takeaways from our discussion: • Finance needs specialized frameworks beyond generic AI metrics • Regulators require clear evidence chains, not black boxes • Traditional statistical models + new AI techniques are key to adoption The full interview is out on Thursday. Subscribe to AI Street to get in your inbox. Link in the first comment. Follow me (Matt Robinson) and AI Street for more on AI + Finance.  

27

3 Comments

While you probably don't have to worry about aliens invading on the fourth of July, you should be aware that text embeddings may expose your private information. Fortunately it's an addressable risk. Our Senior AI Scientist, Joseph Ferrara, PhD, covers how to mitigate the problem on a pdf. Broadly, the steps are: -Extract the text -De-identify the sensitive info -Chunk the resulting text -Embed it in a Pinecone database -Perform some test queries If you want to see more detail and sample code, check out the how to guide the comments. #RAG #AI #embeddings

15

3 Comments

We recently worked with Nirmal Thacker to fine-tune GradPilot, his AI assistant for helping grad student prepare their Statement of Purpose. Our experts created training data that made a huge difference, outperforming the latest Claude models. Check out how we did it: https://lnkd.in/gUWADeEh #AI, #ArtificialIntelligence, #Finetuning #Claude #Anthropic

35

2 Comments

For all you AI and data needs this is worth being aware of.

4

Some thoughts after spending some time with a handful of builders and investors in NYC this past week! 1. New York City is quietly fostering one of the most vibrant and underrated consumer founder communities in the country. This ecosystem is bursting with innovative energy, driven by founders who are deeply embedded in culture and consumer behavior. For those who haven’t tapped in yet, I strongly recommend paying attention—it’s a hotbed of potential that’s primed for growth. 2. Understanding and valuing consumer products demands more than just financial acumen; it requires a deep understanding of culture, trends, and human behavior. Many investors are stepping away, not because the space isn’t viable, but because they lack the nuanced taste and cultural insight necessary to identify and nurture truly impactful consumer products. This gap presents a massive opportunity for those who “get it” to dominate the market with culturally relevant investments. 3. In today’s world, a product or service that lacks a strong community component is missing the mark. Community isn’t just an add-on; it’s the foundation upon which modern consumer brands are built. Doubling down on building and nurturing a community around your product can create a powerful network effect, driving organic growth and fostering long-term loyalty. 4. The next wave of consumer social platforms won’t look like the traditional social networks we’ve become accustomed to. Instead, we’ll see social experiences seamlessly integrated into other aspects of daily life, such as fashion, education, and real-world interactions. This evolution will redefine how we connect, share, and engage, creating new opportunities for innovation at the intersection of technology and everyday life. 5. Simplicity wins — In a world overflowing with options and information, simplicity is the ultimate competitive advantage. Products and services that cut through the noise with straightforward, intuitive designs and clear value propositions will stand out. Complexity might seem impressive, but simplicity is what drives adoption and sustained success in the consumer market.

64

9 Comments

🚀 New Post Alert 🚀 To ASK’EM and beyond (1 of 2) A path to scalable, data-driven decisions I recently came across an article by Mary K. Pratt featuring ASK'EM—an embedded product survey system I created at Indeed to give product teams targeted, actionable insights that directly connected to user needs. ASK'EM was a three-and-a-half-year passion project, so seeing it recognized as a case study was one of those 🤯 moments! Today, I’m building on that foundation at Splunk. We’re using my AAIM UP with U3 approach to deliver products that are Useful, Usable, and Ubiquitous (U3) and aligned with Upward Potential(UP)—creating a continuous loop of Monitoring, Improving, Alerting, and Attributing (AAIM) to keep product decisions relevant and impactful. Interested in how these approaches can turn data into a powerful growth engine for your products? Check out my latest post! Here’s a preview of my guiding principles: 🌟 Build on a durable foundation that targets what matters most 🌟 Align goals and metrics with customer needs and business priorities 🌟 Meet customer needs with intuitive, adaptable products ‘Til next time, I’m Bianca 👉 Read the full post here  https://lnkd.in/eZuvi25B 👉 Read about my approach here  https://lnkd.in/ezpcvjdj 👉 Read the original article by Mary K. Pratt here https://lnkd.in/eUY6EaZx #DataDrivenDecisions #MeasureWhatMatters #ASKEM #UserFirst

11

I’ve aggregated a list of 50+ VC roles in one spreadsheet. It’s updated weekly and can be filtered by location, position, firm, etc. Check out the latest list here (as of June 24th) 👇 https://lnkd.in/dRC3XtjS Tag someone who needs this in the comments 💭 — Want a quick snapshot? Here are a few of my favorite postings: Associate, M12, Microsoft's Venture Fund Link: https://lnkd.in/eNFaaSBy Associate & Principal, Wischoff Ventures Link: https://lnkd.in/gu6Mun7p Consumer Investor, Stripes Link: https://lnkd.in/eZdRk6jV Venture Lead, Struck Studio Link: https://lnkd.in/exGcNHJ5 New Ventures Associate, Live Nation Entertainment Link: https://lnkd.in/e3mBYGCt — I’m creating more tools, guides, and other free resources for Junior VCs to (1) 2x their network and (2) carve their path to partner. If you found this helpful, check out the Venture House website for more 👇 https://venturehouse.xyz/ #venturecapital #startup #recruiting #hiring

17

1 Comment

I often feel like I'm missing out, seeing the high density of startups in NY and SF while being in Boston. But a direct consequence of these networks is the creation of echo chambers. I understand that certain technologies are exciting, but I don’t see much differentiation between the ideas. In particular, I’ve heard a mind-numbing number of startups working on agents, and they all sound essentially the same, with only minor variations. What separates the signal from the noise for me are founders who chose their idea because they have a truly unique angle, driven by their experience or a deep passion for that space. This is what distinguishes founders who are jumping on the bandwagon to pass time and cash in from those who actually have a chance at creating something with staying power. Venture capital needs to incentivize greater exploration of ideas. If you’re building for a problem that already has solutions, then you’re building for the past, not the future. Some may disagree with that sentiment, but my point is, if there are already solutions on the market, you need to be significantly better to gain adoption. And so if you're building to compete, you're not focused on defining the future of your technology. Your focus is shaped by what your competitors are doing. The issue with this scenario is that, right from the start, you're giving up one of the biggest advantages a startup has: the ability to execute on bold risks for great rewards for your vision of your startup. You're playing their game, not yours. I find it hard to digest incrementalism and strongly believe early stage venture needs to move away from convergences of ideas. Instead, it should incentivize expanding the surface area of exploration over idea space, which would maximize technological progress and is the best way to uncover outsized value. #startup #venture #investing #machinelearning #musing

9

In Marketing Mix Modeling #mmm, confounders like seasonality, holidays, or macroeconomic indices can distort the relationship between Ad Spend and Sales. Without adjustment, we might overestimate advertising’s impact. Accounting for confounders “unrotates” the data, revealing the true causal effect—a small step with big implications for accurate insights. This is why I’m creating automatic data import as part of MMM Labs platform. We know this data is crucial to the success of an MMM project, so it should be in the model automatically. Prof. dr. Koen Pauwels, Ernest Baver, Vishal Sharma, Mehmet Ozgun Kurt, Dr. Robert Kübler

30

2 Comments