Anthony Sutardja

And just like that, Waymo announced they have an end-to-end model for autonomous driving. Former CEO John Krafcik states: 1. there is long-term promise for an e2e approach (e.g., interpretable decision rationale is possible), but 2. big challenges remain (insufficient long-term memory to solve complex long-tail challenges, sensor fusion), so 3. it's going to be super hard for e2e-only companies (Wayve, Waabi, Tesla) to catch Waymo, which has the only safely working AI approach in the space & a huge on-going advantage: Alphabet AI adjacency/talent

120

11 Comments

Hot take on a lot of early-stage digital health companies out there: You're not a tech company. Most tech-enabled care companies I've seen over the past two years are only tech-enabled in the sense that they use telehealth as their delivery mechanism for care. There are exceptions to this, of course - companies who build computer vision and applied AI tools to better the doctor-patient experience, companies who build tooling around risk scoring and stratification to enable more personalized medicine, companies who build truly novel intakes to better route patients to the appropriate care, etc. These companies have real technology chops and, if they sell it right to the market, can get credit for this from investors. But if you're truly a telehealth provider, you need to realize now that the goal post has shifted dramatically from 2021 and you're going to be valued at maturity at something like 3x EBITDA. That's profit. Not revenue. Throw a revenue multiple out the window. That's not the world we live in if you're a virtual doctor's office. Think of how that impacts venture math. If you're raising a seed round at $15M post-money (or $12M pre-money), it means you'll have to generate something like $150-200M in annual EBITDA to be a fund returner for your seed investors. Maybe even more with dilution, since telehealth businesses are not traditionally capital efficient and will likely need to raise material additional capital to reach that scale. You need to be ridiculously profitable (probably on $800M-$1B in annual revenue since you're likely going to be a 40% gross margin business). Here's the good news: you can set yourself up well (and probably be a contrarian founder relative to your peers), by talking about EBITDA growth and how you can grow into a fund-returning outcome based on realistic mature market comps. This will stand out for the investors you're pitching, even at the early stage. This is how you can beat the malaise of the tech-enabled care market. Focus on the right metrics and the right type of growth. You don't have to have all the answers (no founder does), but you need to make sure the ship is headed in the right direction. A focus on profitability for "tech-enabled care" (*ahem* telehealth) businesses is a good first step.

26

3 Comments

Open-source models can be a strong lever for increasing gross margins at AI companies, especially those that have hefty cost structures (e.g., enterprise sales orgs that require significant customer success or implementation resources, as well forward-deployed AI services models common in healthcare and life sciences). One of many reasons to maintain a modular and flexible stack with minimal dependencies when building an AI company. More on the importance of flexibility in the AI stack: https://lnkd.in/eA2XATRk h/t Delip Rao #ai #artificialintelligence #generativeai #healthcare

50

6 Comments

As LLM's get increasingly more sophisticated, with OpenAI's o1 model now including reflection before answering, I am increasingly investigating the emergence of autonomous AI Agents. My path here was somewhat circuitous in that I was going through Y Combinator's Summer 24 cohort which had a plethora of AI use cases that were very industry specific. I found that curious in that earlier SaaS offerings should have already automated and/or digitized most of these process, and I couldn't figure out why AI startups would focus on basic workflows such as inventory management and book keeping. As I dove into the matter, with the help of my now indispensable Perplexity, I landed on a very insightful post by Bessemer Venture Partners (https://lnkd.in/gHtQQRha). The thesis is that Vertical AI companies are going to be magnitudes bigger than general solutions, especially as they focus on industries that were earlier inaccessible to SaaS given very high language based workflows. Within that thesis, there was also a mention of AI Agents and how these would come to dominate workflow automation going forward. This neatly tied in with a great conversation between Peter H. Diamandis and Emad Mostaque (https://lnkd.in/g-ccxbma) that talked about 10 billion agents!! I have spent the last few weeks diving into AI Agents with the help of content from interesting voices such as Adam Silverman (Hiring) and David Ondrej. Over the next few months, I will try to create and deploy AI Agents on some basic use cases to get hands on experience on these systems. In the meantime, I will close this post with a link to a video from Andrew Ng, a pioneer in AI systems that provides a short overview of AI Agents (https://lnkd.in/g8v4pUQu). I hope you enjoy the content and wish everyone a productive week ahead.

8

Great insights on the future of robotics shared by Rodney Brooks in this article. A little bit of realism for the hype surrounding LLMs in robotics - realistic expectations are important...

4

Next level AI application from the NewNow team. Check out the article.

11

Outrider just raised $62M to scale up in autonomous yard operations. This is noteworthy because one of the best use cases for autonomous transportation is in closed-loop transportation networks, where there is not a risk of hitting pedestrians. We've seen self-driving tractors in mines, military applications, and now yard operations. Outrider's investors include Koch Disruptive Technologies, New Enterprise Associates (NEA), 8VC, ARK Investment Management LLC, B37 Ventures, FM Capital, Interwoven Ventures, NVIDIA's nVentures, and Prologis Ventures. Some of these firms are strategics, e.g. Prologis, who can help with adoption. Kudos to Andrew Smith and all involved. The future is bright for transportation innovation! #supplychain #transportation #technology #investment https://lnkd.in/eZ9t7Zhy

66

3 Comments

There's nothing like real data to evaluate if a particular type of robot is commercially ready. Last Friday, we received some exciting data about sidewalk delivery robots when Serve Robotics listed publicly on the Nasdaq exchange. As quick background, Serve was briefly part of Uber but was spun out in 2021 to focus on building and operating their sidewalk robots, primarily for food delivery. So what does the recent filing reveal? First, we learned that Serve revenue is only 50% from delivery fees while the other 50% comes from branding partnerships. With Uber and Lyft both investing heavily in monetizing their advertising surfaces, this is not surprising. The percentage of revenue from branding is likely to reduce over time. While revenue from delivery fees should increase linearly with delivery efficiency, branding partnerships are likely more fixed. Second, we learned that Serve is operating an average of 34 robots per day out of a fleet of >100. Based on their claim of 50k deliveries completed since early 2022, this implies roughly 0.4-0.6 deliveries per robot per hour. This rate is well below the efficiency of human couriers but likely includes significant idle and down time and will improve over time. Third, we learned that Serve loses significant money on each delivery. Again, using their claim of 50k deliveries completed since early 2022, I calculate roughly $6 in revenue and $50 in costs per delivery. A loss of $44. While scale will certainly help increase revenue and decrease costs, this is a giant gap to close. With a claim of only ">80% autonomy", a big portion of this cost is likely in remote and on-the-ground support. Note that if you include all of Serve's operating expenses (e.g. R&D, Ops, SG&A, etc.) they actually spent ~$600 per delivery in 2023. Finally, we learned that Serve aimed to raise ~$37m through the listing with a burn rate in 2023 of ~$25m. With plans to invest in fleet expansion (up to 2,000 robots by the end of 2025) and a next generation robotic platform, their runway is limited (<18 months). With unit economics that are upside-down, expansion only accelerates this burn. The best part about Serve Robotics being public is that we get to see more and more of this data each quarter. I'm hopeful that sidewalk robots can be part of the last mile solution but autonomy needs to improve and costs need to come down, quickly. Let me know what you think about sidewalk robots below.

85

8 Comments

WAYMO: THE BIGGEST SUPPLY CHAIN TECH INVESTMENT OF THE YEAR The supply chain technology markets are heating up. In the biggest transportation tech investment of the year, Waymo just announced a $5.6 billion financing. The investment, led by Alphabet Inc. (which owns the majority already), also included Andreessen Horowitz, Fidelity Investments, Perry Creek Capital, Silver Lake, Tiger Global Management and T. Rowe Price. Waymo will use the capital to pursue leadership in the autonomous vehicle arena, with a focus on its Waymo One robotaxi offering. Would you get in a self-driving taxi? Most people don't realize this, but there are already 100,000 weekly trips in San Francisco, Los Angeles, Phoenix, Atlanta and Austin. I predict robotaxis will be in most major markets within 3 years. What other winners are you seeing in the autonomous transportation arena? #supplychain #transportation #tech #autonomous https://lnkd.in/e8HAhReY

42

11 Comments

What a week this has been for innovative companies in Seattle (first Smartsheet and now OctoAI). Today NVIDIA made public the news that they have acquired Madrona portfolio company OctoAI. Partnering with the founders since Day 1 (ok Day -1) has been one of the greatest experiences of my life. Luis Ceze, Jared Roesch, Tianqi Chen, Jason Knight, Thierry Moreau and Carlos Guestrin were all involved in the research and insights that helped create the TVM open source technology at the University of Washington computer science school and then the creation of OctoAI back in 2019 to drive commercial impact with this technology. From sorting out the terms of the original "pre-seed" round over a bottle of Caymus Vineyards wine to helping Amazon Web Services (AWS) understand the dynamics around the open source movement in Applied AI to learning our way in to product-market fit with the BEST inference optimization service available (OctoAI can help customers find their optimal fit of performance, customization, and cost), OctoAI has been the best! The company has helped set the table for AI-driven innovation for the years ahead and they will be a perfect compliment to the many capabilities at Nvidia. Luis, his leadership team, the board (including Carlos Guestrin Lip-Bu Tan and Mike Dauber) and so many "friends of OctoAI" know what a motivated, talented and authentic group of people are associated with this company. It is inspiring to see an acquisition at a total deal value of over $250 million in this highly dynamic and fast-moving phase of AI disruption. It is even more exciting to anticipate what will be possible with innovations powered by AI models and related technologies in the years ahead. Madrona has been investing in Applied AI for over a decade and we truly believe that we are just getting started! One of the hardest judgement calls as an investor and board member is to assess the benefits and risks of selling a company at a key inflection point in the journey. Nvidia made a compelling case that the impact for the OctoAI team and technology would be maximized within the market leader for AI infrastructure. I will be rooting every day that they all maximize this impact in the months and years ahead!

748

54 Comments

Waymo is the best tech product I tried in years, and I can't wait for humans to be banned from driving. What's Waymo? Waymo has been at the forefront of autonomous vehicle development since its inception as the Google Self-Driving Car Project in 2009. The core of Waymo's technology lies in its sophisticated combination of sensors, including LiDAR, cameras, and radar, integrated with artificial intelligence to perceive the environment accurately. This technology allows the vehicle to make quick decisions in real-time, such as identifying pedestrians, bicyclists, and other vehicles, understanding traffic signals, and adapting to unexpected situations like road work or detours. Over the years, Waymo has expanded its testing grounds from a single city in California to multiple locations across the United States, including San Francisco, Phoenix, and Detroit. In addition to testing, Waymo has launched pilot programs like Waymo One (The service I tried), a self-driving taxi service in San Francisco, Phoenix, and Los Angeles. You can use the app to request a ride in a self-driving vehicle. This service marks a significant milestone in commercializing autonomous technology, offering a glimpse into the potential future of urban transport. Additionally, Waymo Via focuses on goods delivery, utilizing the same technology to transport products efficiently and safely, which could revolutionize supply chain logistics by reducing costs and improving delivery times. I am grateful to Amjad for the access to Waymo One. Which city in the Middle East will be the first to welcome Waymo? I would say Dubai, what about you? #selfdriving #waymo

22

9 Comments

NVIDIA's Jensen said "ChatGPT democratized computing, Llama2 democratized generative AI" ...and now Llama3 is the next level 🔥 It's fascinating to watch Meta's strategic moves. With Llama3, they prioritized training on a massive 15T token dataset to pack all into a lean 70B param model, instead of building a massive model. This allows Llama 3 to match trillion+ parameter models like GPT-4, but at 1/10th the compute, storage and inference costs! 💰 This technique was published by Google DeepMind a few years ago https://lnkd.in/gx7VU8aA Meta is not a dark horse in AI anymore. They might be the top dog. https://lnkd.in/gRRJYuxt #llama2 #llama3 #llm #chatgpt #gpt4 #ai #ml

16

5 Comments

This is f*cking hilarious: TuSimple's pivot from self-driving trucks to AI animation and gaming! 🤖🎮 Right now, everyone’s betting on humanoid and general-purpose robotics startups. But after 5 years in self-driving tech and building Kopernikus Auto, here’s the reality: most humanoid robotics companies will be dead. In the self-driving world, we’ve only seen a few real winners: Tesla, Waymo, and Comma.ai. Why?  - Humanoid robotics isn’t a hardware problem; it’s a software problem. - You either have infinite data, cash, talent, or a clear use case—or you won’t make it, no matter how much you raise. From over 100 companies, almost all in self-driving tech are dead or miles away from profitability. ✨ Don’t get distracted by the appeal of humanoids. Focus on real problems you can actually solve. What are you focusing on? 👇

13

1 Comment

Waymo's recent doubling of its paid ride volume is an impressive business result. But what does the feat say about the state of the self-driving industry? And how does Waymo stack up against self-driving rivals like Tesla, and ride-hailing giants like Uber? We dive into these questions on the latest episode of TWIST

23

3 Comments

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