Charlie Kingston

At Pavlov, our goal is not just to share what we’re building, but also to provide insight into the thinking behind it. Niall Kelly did an amazing job of that in his most recent post, “More Metrics, More Problems” (linked below), where he explores the challenges of dealing with too much data. In my conversations with customers about their tech stacks, they’re often initially concerned about both the quality and volume of their data. The truth is, by focusing efforts on small subsets of data (sometimes even just one data point), you can deliver real business value. Niall and I thought it would be valuable to dive deeper into this topic in the conversation below. Let me know what you think.

31

1 Comment

If you're wondering how AI will be monetized, and how we'll come to pay for it, this article by Abde Tambawala and https://lnkd.in/gYJxgPSS is worth a read. If you're building an AI-driven product and are thinking about how you'll price and package it, this is a must read.

7

1 Comment

Entrepreneurship is a tough but incredibly rewarding path! A big thanks to Outward VC for highlighting the story behind Capsa AI and my own journey. Excited for what’s ahead! Check out the post below 👇 #Entrepreneurship #PrivateEquity #AI #CapsaAI #StartupJourney

41

4 Comments

🎯 AI isn't just for consumer products - it's transforming how teams work behind the scenes. During Google's latest earnings call, their CFO, Anat Ashkenazi, highlighted AI-driven efficiency in internal processes as a key focus. When a tech giant is looking to AI to streamline their operations, it's a clear signal for where the future of work is heading. Questions for finance and accounting leaders: ❓What manual processes still eat up your team's time? ❓Are you positioned to lead this transformation, or will you be playing catch-up? The future of finance isn't just about analyzing data - it's about transforming how we produce and validate them. #AI #FinancialReporting #CFO P.S. If you're looking to upgrade your financial reporting, check out what we're doing at InScope! https://lnkd.in/gQ_wZqAD

15

Founder-mode vs Manager-mode Paul Graham’s published an essay yesterday on 'Founder Mode' and it has been all over my twitter and linkedin timeline. It is a must-read for founders but I also found it a bit amusing. He explores how scaling advice meant for managers often clashes with the instincts of a founder. It’s a great insight, but one that many of us building businesses outside Silicon Valley—or without a formal business school background—already understand well. We've been navigating the complexities of hands-on leadership every day, often without the luxury of traditional advice. I feel as a founder the best approach often comes from our own experiences, instincts and our guts and there is no playbook to success. https://lnkd.in/d9GDKx2J #founder #startups

2

One of the best blogs I've come across on quantization, with outstanding visualizations. Highly recommend a read for anyone looking to understand LLM quantization in more depth! 📐 - https://lnkd.in/e_WTgq6x At Felafax AI (YC S24), we are building Infra to make full precision multi-GPU fine-tuning easier. #AI #Infra #LLM

16

3 Comments

The Four Horsemen of Customer Churn Our data scientists have combed through mountains of unstructured customer usage data to crack the code on proactively identifying accounts that are a churn risk. After analyzing thousands of signal combinations, we found that four key indicators—Budget Issues, Unhappiness, Value Issues, and Urgency—are the ultimate predictors of revenue risk. Nearly every B2B tech and services company sees the same pattern: when these signals align, it’s time for action. Hold on, what is unstructured usage data? It’s the raw, untamed data that tells you what customers are *really* doing and saying—not just what they’re willing to admit in a survey or conveyed by numbers of daily average logins (also critical but lacking context). Here are the harbingers of risk; when combined, they are what the team needs to act on right now. 🧯 1️⃣ Budget Issue: This signals a customer struggling to justify the cost, possibly due to tighter budgets or a perceived lack of value. 2️⃣ Unhappy: Customer dissatisfaction can stem from unmet expectations, unresolved issues, or lack of engagement. 3️⃣ Value Issue: If a customer doesn’t see the ROI, they’ll start questioning the worth of your service. 4️⃣ Urgent: An urgent flag indicates an immediate problem that requires rapid action. They are expressing a need to engage with a teammate now. 

28

One year ago, we started Epsilla (YC S23) with a big dream. Today, as we celebrate our first birthday, I want to share the story that brought us here. Many people ask why I left my job as a senior director at TigerGraph, took a huge pay cut, and founded my own company. Since I was a kid in elementary school, I've watched the story of Bill Gates building the Microsoft empire, and dreamed of one day building my own company. This dream grew stronger when I came to the US in 2014 pursuing my master's degree at Cornell, and applied to Y Combinator with my roommate, Hao Ma, with a simple anonymous social app idea. Even though it didn’t work out then, it planted the seed in my heart. During my time at Cornell, I met 4 other entrepreneurial students, Landice Gao, Xiaoyan Wu, Jiaqi Su, and Kejia Tao, and started a company focused on indoor navigation. I wished it could have gone further, but as an international student, I needed a job that could sponsor my working visa to stay in the US. So I joined TigerGraph at 2015, a pre-series A startup at that time. Since day 1, I was trying my best to grow as fast as I can. At TigerGraph, I learned a lot from industry leaders like Amol Ghoting, Li Chen, Mingxi Wu, Ruoming Jin, Yu Xu, Like Gao, Tim Zhang, Songting Chen, Shuo Yang, and many other talented colleagues. Most importantly, I met my future co-founders, Jing Qin and Eric Y., and our future advisor and angel investor, Jay (JieBing) Yu, PhD. When ChatGPT launched in November 2022, it blew my mind. I thought such advanced technology would appear in 2045 (according to "The Singularity Is Near"), but it was already here. However, I quickly realized that large language models (LLMs) didn’t have memory and their knowledge was fixed. Our friend Weimo Liu mentioned vector database to us, and I saw the huge potential of combining AI with data. This led to the birth of Epsilla, aiming to build the data and knowledge foundation for LLMs. Becoming a CEO from a technical background has been tough, especially for an introvert like me. I am so proud of my growth so far. In the past year, I’ve talked to more people than I have in my entire life. I’m grateful for all the interesting people I’ve met around the world, which wouldn’t have happened without starting this journey. Looking ahead, I’m very dedicated to the success of Epsilla. Our journey is just beginning, and I’m excited to see what the future holds for us. Thank you to everyone who has been part of this journey—our customers, partners, team, investors, and supporters. Your belief in our vision keeps us going every day. Here’s to one year of Epsilla, and many more to come! 🚀✨

105

32 Comments

Great to see this paper by LinkedIn showing an uplift of 29% in using KGs in RAG for a customer service use-case. The main value I see the KG adding to the system here is around query reformulation into structured categories (i.e. specific tickets), automatic categorization and classification of issues for ancillary flags (Priority/Root Cause), and structured retrieval of ancillary data ("Has_Steps_To_Reproduce") to ultimately return related but non-semantically similar information. It's interesting they still had to use semantic similarity for the ticket identification step and there's an important caveat here the authors mentioned. Dynamic and automated schema extraction are future areas of work. Link to paper: https://lnkd.in/emqh3bUC

149

12 Comments

Interesting answer from AlphaSense CEO Jaakko (Jack) Kokko on why they acquired Tegus - all about access to proprietary data, in this case the transcript library on private companies. You recently acquired a company called Tegus. Was this about having more proprietary data to feed into your models? First and foremost, it was about providing our customers with the best possible intelligence on the market, especially on private companies. A couple of years ago, we acquired a company called Stream, the No. 2 player in what you might think of as a library of interview transcripts between investors and experts on a company — their former executives, competitors, customers, partners. It allowed us to produce information that you can’t access anywhere else, and our customers loved it. Tegus is the market leader in that sector and was the natural next step for us. They have the world’s best library of private company content, and we immediately wrapped our search capabilities, our generative AI, our monitoring capabilities around their content and made it much more valuable. We now have the broadest coverage of insights on both public and private companies, allowing our AI to provide even better answers to our customers.

161

8 Comments

I am often asked about the “hardest thing” in running a start-up Hiring great people is hard, fundraising and managing a cap table is hard, competing with very well funded incumbents is hard, finding product-market is hard (probably the hardest one on that list if you ask me..) - a lot of founders talk about these. But for me, especially on the GTM side of things, the hardest thing is the insane power imbalance between a start-up and pretty much everyone it’s selling to. No one talks about it but it exists and it gave me more sleepless nights over the last 3 years then any other problem. Let me explain. As a devoted Sandler fan, I live and breathe by “equal business stature”. Your time is valuable and so is mine. Before investing into anything, get clear commitment from the other side. Walk away from the relationship if effort is not reciprocated. This approach served me incredibly well in my 9 years in investment banking, sales and investing but it sadly doesn’t quite translate to early stage start-ups. The last 2 years have shown us all that you can’t build a business out of selling to other startups. If you want to survive, you have to go after larger companies, partner with established service providers and generally swim with much bigger fish. They have thousands of employees, you have 15. They have a brand name, you are a new kid on the block. Most importantly, they have time, they can afford to wait a few quarters… you need this deal signed yesterday. Everyone knows it. They can afford to f*** around, you have to play by their rules if you want a shot at working together. I WISH walking away from a deal after someone reschedules a meeting for the 3rd time or wants a trial extension was a possibility - but in reality it’s not. One large customer and one successful partnership can change the game in the early days of a startup. So you end up over-investing and giving more than your fair share, violating the “equal business stature” rule many times over, but I have yet to find another way to open doors and actually get a shot at the goal. Do I still believe in “equal business stature”? Absolutely. But there is this weird corner of commercial operations, called the zero to 1 game, where rules that apply to 90% of situations cease to exist. 

38

9 Comments

Brilliant illustration of all the benefits that come with a Revolut Business Account. If you would like a better understanding of how we can help your business grow, let's connect!

5

Wrapping up my trip with a visit to Google in Cambridge and some reflections on why FAANG is not my cup of tea: First, it's worth mentioning there's a lot to LIKE about a big tech career - Get paid a LOT of cash, and stock you can sell for more cash - Solve problems at a scale larger than most startups, by orders of magnitude - Work with generally smart, self-motivated people, since the hiring bar is high - Fun office spaces where you can work out of a teacup if you'd like - If you get bored you can move around the org and the world I wouldn't call such an opportunity unenticing. If I had chosen offices with teacups at the start of my career, I'd probably have had a lot of fun, learned a ton, and dig the perks. And there might have been a version of events where I joined successively great teams, found the right managers and projects, and kept having fun for a long time to come, while earning a ton of real money. But there was also a very likely path (which I see a lot of folks go down) where - bureaucratic bullshit fills the day - politics dictates how products are built - the work gets so uninspiring that coasting becomes more attractive - hit some glass ceiling in career growth because of bad management - get sick of working for other people who work for other people who work for other people So it's not OBVIOUS that I'm better off for having gone into early stage startups and now starting my own. But, I'd say I'm content with my path - I choose what I work on and who I work with - I never go into work dreading a conversation with someone difficult or toxic - If I get absolutely bored with what we're working on, we just go pivot - I never have to fight for 'credit' for my work. - When I see bullshit creeping in I can cut it right out - I never have to ask for approval to do something - I'm answerable only the The Market (and my cofounder? 😬) - If I want to go to somewhere different / nice / fun to work I can just go fly somewhere and visit my friends anyway. Would you trade what you're doing for a FAANG job with a fat paycheck? (Or if you already have one and LOVE it, please tell me what I'm missing out on?)

44

12 Comments

Just sent out the latency newsletter for April. Noticing some really interesting trends. 🏆 Claude Instant regains the top spot after losing it for only a single month 🐢 Claude Sonnet + Haiku both saw a major increase in latency (~35% on average). This pattern of a slow down in performance the month after launch happened to GPT-4-Turbo as well. Could be related to increased adoption as people had more time to test and implement. 🐇 A model that I really like but doesn't get much attention, gpt-3.5-instruct, saw the biggest gains ( >50%) Plus a whole bunch more Get access to all the data via the link below. One email, once a month, packed with actionable and data-backed recommendations.

9

3 Comments

Getting computers to talk to each other is hard. 🗣️ There's a lot of hype about LLMs eliminating structured data altogether, but that won't happen because, fundamentally, natural language lacks *precision*. Here's why I think data standards aren't going anywhere: 1) Precision 🎯 - Natural language is inherently ambiguous (unless you're speaking Esperanto). We need data standards for the same reason we use mathematical notation: to remove ambiguity. When someone writes "5/10," we need explicit context to know if that's a date, rating, or fraction. 2) Monologue vs. Dialogue 💬 - When humans book international flights, we naturally engage in back-and-forth: "Is that departure time in local or destination timezone?" With wire transfers, we clarify: "Is that account number IBAN or SWIFT format?" But machines typically communicate in one shot: request/response. Even with LLMs, network contraints reinforce this pattern. 3) Validation ✅ - Because there's no dialogue, machines need to get it right the first time. Data standards enable this through validation: Is this price in dollars or euros? Is this product identifier UPC or EAN? While this can get pretty wonky, it determines whether a transaction succeeds or fails. The evidence is already here. Even ChatGPT uses function calling to make its outputs more deterministic (https://lnkd.in/gSJBreWE). And interfacing with LLMs themselves requires structured protocols, as shown by initiatives like the Model Context Protocol (https://lnkd.in/g9ZQJ_rg). The future looks less like LLMs replacing standards, and more like LLMs-as-compilers: translating human input into precise, machine-readable formats. Thoughts? When have you seen ambiguous data cause real business problems? #AI #FHIR #Integration

26

3 Comments

Every week with Oliver Strijbis, we scan the different prediction markets and forecasting tournaments. Our goal is to look at where probabilities changed the most, and why. We package all of that in a newsletter : Forecast Brew. We just released our latest issue 🌏 South Korea: Will President Yoon Suk Yeol face impeachment by year-end? 🇷🇺 North Korea-Russia: Could North Korea deploy troops to assist Russia in Ukraine? 💡 Nord Stream 2: Is the pipeline set to reopen by 2025? 🐦 Bird Flu: Will the pandemic risk rise? 💰 Fed Rate Cuts: Will December bring another rate cut?

6

if you know, you know xAI announces 'colossus' 100,000 h100s. sam i am whatev you say i am (alt)man aims to spend billys. so. what's the story morning glory. it could be just me. it could be following the (ai) news stream online, on X. it kinda appears that whenev elon makes a statement. toute suite. there is a reply from OpenAI via bloomberg. is it me. or is this part of the nerd drama. tit for tat. either way on the back of NVIDIA hitting it outta the park with it's valuation. the other 2 indicators (amongst many) appears to be up and to the stratosphere. which creates the resulting fomo. with the rest of the field anticipating that all this good news trickles down to (ai) chatgpt grok llms whatev being implemented and in fact performing as advertised. there is another factor which has to do with the (un)natural implementation. adoption. of these clusters. the energy. the queue. the models. the data. the resulting outcomes advancing varying industries. first the creative in the form of images, music, video. then in areas as in biology, pharma, health. then perhaps in areas which become it's own new marketplace that appears as if from no where. and so it is. we are in a market which is gun ho. ho ho ho. in time for christmas those NVIDIA gpus will be cranking out the training outcomes. there is a lag when this stuff is in training mode. it's weeks and months depending on the mixture of tokens, the model, complexity of other models. and of course the output that matches as if by magic. there is so much going on at the same time we are watching intently for all the great news. a rising billy tide that raises all millys to billys. exciting stuff. as we watch the tit for tat. goes click clack. train. infer. and back again. source: https://lnkd.in/eNuC9dtW source: https://lnkd.in/g9aBN2Qf < . >

2

4 Comments

Hiring and working with excellent people is what makes starting a company so fun

16

✨Will companies need to build Eval teams to constantly review their AI Agents? 📋 I’ve recently been digging deep on how to make sure a couple of agentic solutions I’ve built continue to perform well My 2 concerns were: 1) How do I know if a new prompt improves or breaks my agent? 2) When Anthropic / OpenAI make changes to their underlying model, how does that impact my agent? Note that these were small personal agents with very little risk so I wasn’t even thinking about security, privacy, safety, ethics, etc. This led me down a bit of a rabbit hole 🕳️🐇 The core problem is that none of this is black and white… you’re evaluating a multi-dimensional spectrum of different shades of grey 🤯 1️⃣ As a result, the first thing I realized was that it’s actually a huge undertaking to just maintain your solution’s integrity… even a change in the foundational model version can impact your agent 2️⃣ The second thing, is that if you’re planning to build a custom model (or even fine tune on top of a foundational model), the problem explodes in magnitude 3️⃣ In a world where a company has a portfolio of in-house and 3P agents, this could quickly become very challenging to manage, especially for a startup / SMB Extending Jensen Huang’s “HR under the CTO” analogy, that’s an awful lot of time doing agent reviews! While 1️⃣ and 2️⃣ clearly sit primarily in the responsibly of the CTO, it’s less clear who is responsible for evaluating 3️⃣ (especially 3P agentic solutions) There are Eval frameworks that have been issued by the likes of Anthropic and OpenAI… but you need to be a data scientist to be able to use these. My key takeaway was that if you’re buying an agentic solution, there are probably 2 different approaches you can take: 🔹 Leave the technical stuff to the vendor and just focus on monitoring outcomes… this has risks! 🔹 Work closely with the CTO, and build a process to programmatically evaluate the solution over time… this requires a larger org (but there may be solutions that exist) My other takeaway is that this is probably the first Org change that AI is inflicting on a business… … The CTO will need to build out an Eval team responsible for managing all the agents and models in a organization #AIAgent #Eval

3