Xuhui Shao

AI memory is fundamental for conversational AI and AI agents. It is critical for building personalized experiences and improving reasoning. Check out Letta’s course to understand how to use AI memory to enhance the customization and quality of your AI app.

15

Sharing a sanitized version of the mu ventures Q2 q'ly LP update. Included musings: 👉 SEOs shrinking impact, and the rise of LLMO (large language model optimization) 👉 examples of the Commerce AI application layer coming to life 👉 a modern fund strategy that accepts companies require less capital, time, and people to trigger the power law 👉 Mu's guide to investing in AI 👉 all 30 companies in the Mu-niverse (and several others we admire) Founders, VCs, LPs - if you like how we think and suspect we can partner, hit us up! https://lnkd.in/dTJ2zjav

52

7 Comments

Breaking down our latest newsletter section by section, starting with our expanded focus in Deep Tech: --- We have built expertise in the robotics sector, and will continue to double (and triple) down on investing in this category, but we have been increasingly finding that robotics is naturally steering us into two exploding adjacent areas: Aerospace, and Advanced Manufacturing (which we define as industrials, semiconductors, materials science, and energy production). Stepping back, this expansion makes perfect sense. Robotics is an inherently horizontal platform - meaning it can be applied productively across multiple industries. This quality, paired with the fact that robotics has long shown the highest TRLs of any deep tech technology, makes it a kind of “gateway” to other sectors in the hard sciences. Now with the recent breakthroughs in AI for robotics, growing semiconductor demand, and the resurgence of aerospace and defense tech, we are seeing a dramatic uptake of robotics and relevant technologies in both advanced manufacturing and aerospace. As an added bonus of the expansion, many of the founder (and funder) networks across these landscapes have natural overlaps and linkages. For AlleyCorp Deep Tech this means that we can be extremely helpful, knowledgeable, and engaged across all three sectors right away, and continue to be the best-in-class investing team across these categories.  --- More in our newsletter: https://lnkd.in/ewiTuYq6

10

I’m thrilled to be announcing that Notion Capital is leading the $15m Series A investment in Cogna today. In the industry, we like to think that software has been eating the world. But when you look at the data a better description would be more of a nibble. Depending on what source you look at the software industry represents between 5-8% of global GDP. Contrast that with the services industry that represents between 60-70% of GDP and worth around $25 trillion on a global basis. Most services have been out of reach for the software industry. They have either been too complex or too specialized for off the shelf software to access in any economically viable way. But that’s beginnning to change with GenAI leading to the emergence of ‘service as a software.’ Cogna is at the forefront of this opportunity - building precision software at scale and delivering huge productivity gains for traditional industries. We know Ben Peters & Lars Mennen well having backed their previous company Five AI that was acquired by Bosch in 2022. And we’re very excited to be backing them again together with Hoxton Ventures & Chalfen Ventures Read my full blog post on why we invested here. Radu Bozga Ben Peters Lars Mennen Bryan G. Mike Chalfen Hussein Kanji Kirsten Connell https://lnkd.in/eh8mhcJb

201

13 Comments

User feedback is always important. Tools like Amplitude and Mixpanel are excellent for capturing user experiences and improving products, ultimately boosting satisfaction and helping businesses stay user-centric. When it comes to AI, improving models is all about learning, and RLHF plays a big role in optimizing performance over time. With LLMs, traditional deterministic evaluations don’t work as well, and RLHF is typically applied only at the foundation model level. A good example is ChatGPT: its performance improved gradually as they collected data from real usage aka interactions. While users can give explicit feedback, most people don’t tend to do it. That’s why ChatGPT sometimes offers two response options to nudge users into providing feedback, which then helps fine-tune the model. However, traditional RLHF doesn’t apply well to the application layer of LLMs, such as in RAG, agents, or prompt engineering. But there’s potential to rethink this. Combining RLHF with tools like Mixpanel or Amplitude could create a more effective approach to improving LLM applications. Placing users at the center of evaluation and optimization is essential — they’re the ones who know their tasks and needs best.

41

Batch processing for LLM inference activity: "One sign of that is the rise of batch processing. Today, popular consumer AI products like ChatGPT and Perplexity provide users with near instantaneous responses, otherwise known as real-time inference. However, businesses don't always need immediate responses and are often willing to wait hours, days, or even weeks for responses, as long as they don't have to shell out as much money. Founders of cloud and inference providers have told me there's growing pressure from business customers for this kind of flexibility, otherwise known as batch processing." -- Nathaniel Whittemore on "The AI Daily Brief" (with much deserved shout-out to his inspo from Stephanie Palazzolo at The Information) I fall into this bucket as we have a lot of backfilling to do as we're building out the underlying data in our service, and some of that could just as well be done in the wee hours of the night. The new stuff flowing in I want processed ASAFP, but these backfills—I'd mos def take more cost-efficient delayed batch processing. There's also use cases like needing (near) real-time inference for the first inference pass for timeliness / contemporaneousness value, but then a second pass slow-time inference for wider classification exercises over an ever-growing data corpus. I could see us mixing Sonnet & Haiku models in both cases. Just a heads up to Anthropic Mike Krieger as you contemplate this, along with Amazon Web Services (AWS) Jeff Barr Mark Roy Swami Sivasubramanian on the Bedrock side of things...and I don't mean the "batch" mode that doesn't seem to have any real value other than helping AWS manage its own rate limit challenges. (DeepCast is using Bedrock.)

11

2 Comments

One of the things we see quite often at Grit Capital Partners, is a newly minted CEO. We see a big difference in the transition to CEO based upon the historical lane the CEO is coming from. There are different new muscles if you are coming from Product, Engineering, Finance, etc. In this article, we go deep on the CRO to CEO transition, which includes some thoughts from Rosie O'Meara, a particularly fine example of someone who has made the transition quite well. What were your biggest transition moments? https://lnkd.in/eEB-K2ZU

79

6 Comments

This is just incredible. Neal Ghosh's demonstration of how #VentureIQ takes deeply technical and complicated content across a range of industrial use cases and ages/stages through an AI-generated conversational podcast (a la #notebookLM) is really mind-blowing. Check out his post, below, and if you haven't seen VentureIQ, yet, let's get you set up to learn more. You can visit VentureIQ.ai or DM any of us over here to book a demo!

2

Tribe AI just raised our first external round of capital from Bryce Roberts at Indie, with significant participation from our customers, partners and network of top AI engineers. We’re excited to share this milestone in TechCrunch (thanks Rebecca Szkutak for sharing our story!) along with some long overdue updates:  💲 Raised $3.25M to keep up with demand and build the platform to enable service delivery at a tremendous scale 🚀 8 figure revenue run rate, more than doubling year over year (and accelerating) 🏛 Enterprise customers from the Fortune 20  🧠 500 top AI engineers from places like Google, Tesla, Meta and OpenAI As a refresher, Tribe is an AI-native services company that helps Enterprises build best in class AI products. Our team is augmented by a talent network of over 500 top AI engineers from places like Google, Tesla, Meta and OpenAI who bring both industry and deep technical specialization to our project deployments. These are the people who have built the infrastructure that is powering the AI wave and know it better than anyone. After bootstrapping Tribe AI since 2019, we decided to raise this capital because the market opportunity was simply too big. Even as we were reinvesting our profits back into the business, we still saw more demand than our capacity could support. We’re grateful to have met Bryce and have him believe in us and the opportunity we saw. We’re also extremely proud to have the major stakeholders in our business (our customers, partners and network of top AI engineers) invest in the round as well. We’re on a mission to help Enterprises realize the value of AI for their business, and we’re building a category defining AI company in the process. Read all about it 👇 https://lnkd.in/dYRDQDjp

348

87 Comments

I started Early Stage Labs because I kept seeing the problem over and over again - seed and series A startups need support in certain functions but don't have the budget/need to bring on somebody full time. We signed our first customer (Modicus Prime) 48 hours after meeting them on what was supposed to be a customer discovery call. Rebuilding their model and getting better visibility into the financial health of the company was a "top 3 problem" for them as they prepared to raise their seed round and as circumstances were demanding they become more disciplined operators. I'm incredibly thankful to Taylor - a brilliant founder (check her out!) - for trusting in a new (or at that time nonexistent!) company. That was a real lightbulb moment for me. We could really delight customers and be super valuable to them in their early days, probably up through their Series B. I kind of stumbled into building this business earlier this year, and I've slow played it a bit as I've wanted to stay very close to customers. "Do things that don't scale." We have learned so much. I'm so thankful. We have a legit chance (~30%) to close this year at $100k in MRR. It's a pure services business, so margins are much lower than the technology companies I've built before. But it would be really damn hard to build a technology business to $100k in MRR just 10 months from your first customer discovery call. I'm incredibly proud that that $100k MRR will represent the fact that we're making several startups better versions of themselves. The thing I'm most excited about on this business is that our customers are some of the people I admire most - founders of early stage companies. You have to be a little bit twisted to go start a company, and that's exactly the people I love spending my time with and serving. If you're on that winding journey and you feel like you need support around finance or growth, we're here for you. Cheers to founders. And cheers to boring services businesses.

59

9 Comments

If we want to see more diverse top companies like NVIDIA, we need to pave the way at the earliest stages—venture capital is that on-ramp. It was a pleasure to join Emerging Manager Office Hours hosted by Venture Forward earlier this week. Thank you to Rhianon Anderson and Maryam Haque for organizing such a value add event! A few (other) big takeaways to share: 👉 Distinguish the Curious from the Serious: Focus your time on LPs who are truly engaged. There are many LPs out there still excited about and committed to the VC asset class. 👉 Showcase Portfolio Performance: With liquidity tight, highlight KPIs, revenue growth, and the value your companies are building now to position for success when markets rebound. 👉 Clearly Differentiate: Provide concrete examples of how your approach stands out and drives unique outcomes, i.e., case studies. It was a pleasure connecting with friends and meeting inspiring next-gen GPs shaping the future of venture capital: William Crowder (Aperture® Venture Capital), Joshua Weed (Veteran Ventures Capital), Sonam Velani (Streetlife Ventures), Marcos Gonzalez (VamosVentures), Erica Duignan Minnihan (Reign Ventures), and Jenny Xiao (Leonis Capital). Here’s to collaboration, innovation, and creating lasting impact together.

55

5 Comments

💡 Why do we need so many evaluation tools for LLMs? As engineers, we build 'production-ready' LLM products using these metrics. But what happens next? How do we maintain control and ensure reliability? At Feedback Intelligence, we’ve crafted a cookbook to keep your LLMs reliable and aligned with user expectations. 🍲 📖 Give it a read and let’s chat!

16

What do bio and health applications mean for the next wave of AI -- specifically about compute? (I'm talking to you NVDA longs!) LifeX Ventures is hosting a super roundtable with some serious techbio leaders tomorrow along with Analog Century Management's Val Zlatev -- come hear Thomas Clozel Sean McClain Chetan Patel at 11am ET For the survey results based on asking 100 techbio CTOs and CEOs, join the call. Here, my own speculations after wrapping up the report: -- Amazon needs to have a GPU program on warp drive right now -- Nvidia needs to have a data center / cloud business in the works -- Edge AI chips are about to be everywhere -- in your watch, car, sneakers, thermostats, gene sequencers, spectroscopes -- All this compute needs way more data: expect a boom in sensors of all types -- Remember human genome moonshot that sequenced one person once? Now you can sequence your every cell one by one. Expect that for protein folding (AlphaFold was that moonshot) and other simulation uses, a full "twin" model of you, cell by cell -- Google has no angle here -- worse chips, worse cloud, worse models, worse phones / edge devices -- Specialty silicon for unique use-cases, lower-power, lower-memory are surely coming fast - like back in the day for mobile phones when Qualcomm, Broadcomm, Infineon, Mediatek, etc duked it out.

26

5 Comments

Insightful breakdown from my great Partner Omri Amirav-Drory on the simple factors that really separate the top 1% of investors from the rest, and how more operators should become investors! https://lnkd.in/gEiZdT2h

16

3 Comments

Ross Haleliuk posted the fantastic article below. This is the world where I live day-to-day. At TechOperators, we invest mostly in early-stage cyber, but we do things a little differently, and we believe there are ways to invest successfully outside of pure Power Law math. The article argues that many security problems are too small for VC. I agree. I often try to convince bootstrapped founders not to raise venture because doing so can turn a successful, slow-growing bootstrapped company into a failed venture-backed company because, despite a large infusion of capital, it couldn't double every year. VC is not monolithic--not by stage, strategy, or style. Venture is often equated with "Tier 1 Venture". Ross argues that VC is not always great for early-stage cyber--and he is right. Bootstrapping AND VC work when incentives are aligned. Does it work for an early-stage VC with a <$200M fund to invest in several early companies at reasonable valuations, setting up the conditions for reasonable exits that pay off for both investors and founders? Yes. Does it work for $800M funds investing in seed stage at $100M+ valuations? Well, that depends! Power law says it does (for VC), but the unfortunate externality is that these rounds destroy companies and founder equity more often than not. There is a role for patient capital in this ecosystem to fuel successful companies that retain exit optionality as they scale--driving exit value for both founders and investors.

35

2 Comments

Was chatting with another investor today on a call, and mentioned Tim Urban's Wait But Why Blog post, "The AI Revolution" and how prescient it was. Given that it was written in 2015, decided to flex AI's latest muscles to give it a new venue: a podcast. Podcast here: https://lnkd.in/gW8Dh89U Pretty easy to do: Step 1: Save the posts as PDFs. Step 2: Upload to NotebookLM, have it create a podcast. Step 3: Upload to Descript, give it captions. Step 4: Export, publish to Youtube. Total time to create the podcast: ~25 minutes. To generate this podcast and publish it even a few years ago would have taken >8 hours. Love the future (Original posts: https://lnkd.in/gW-PQXE7, https://lnkd.in/gg-nRvxz) #AIRevolution #GenerativeAI #WaitButWhy #TimUrban #NotebookLM #AIContentCreation #Podcasting #Descript #ArtificialIntelligence #ContentFlex #ProductivityHacks #FutureOfAI #AIinMinutes #TechEfficiency #AIandCreativity #YouTubePodcast #InnovationInContent #vibecap #vibecapital

5

During yesterday's LibreQoS APNIC webinar, I posed the question: won't more bandwidth solve these problems? As Herbert Wolverson said, yes, more bandwidth is good, but, still these problems remain if queueing is done incorrectly. Here's my take on why bandwidth alone is not as good as bandwidth + good queueing policies: Generally it was believed that 'data is important, so don't throw it away; hang on to it and send it later'. In practice, this has proven to be suboptimal as two issues may emerge: 1) latency increases for some flows due to heavy demand from other flows using the same bottleneck link, 2) even a single flow can have excessive latency due to aspects of typical TCP behavior (referred to bufferbloat) when the buffers grew large enough that the data being buffered was retransmitted anyway. It turns out that not all data is equally important. Active Queue Management is the art of deciding priorities, both in deciding what data to throw away, but also in allowing some later arriving data to be transmitted ahead of data in another connection that arrived before it. These problems have been studied, and good solutions have been found by using certain queueing policies in routers and switches, referred to as “fq_codel’ and “cake”. These track the different flows not by classifying the data, but by watching the behavior. Flows that send relatively little data (DNS lookups), or at a measured pace (video chat) have priority over flows that send a lot of data as quickly as possible (App and System updates, Video and ISO downloads).

10

5 Comments

Go Gaurav Sharma and Capitalize! If you have a 401k from a previously employer, check out Capitalize.

26