Lance Co Ting Keh

Hot off the pre-print server – an important publication authored by the Atropos Health team exploring the performance of LLMs in answering real-world clinical questions. LLMs excel at summarizing and contextualizing existing literature, but cannot generate de novo medical evidence today. The results of the study highlight performance gains attributable to injecting novel, relevant, and up-to-date medical evidence into LLM-powered clinical decision support systems. Brigham Hyde Nigam Shah Saurabh Gombar Read the study here: https://lnkd.in/ecuU25Pq

49

2 Comments

Nearly half (if not more) of YC's F24 healthcare startups are building AI agent tools for doctors. Using agents to answer phones, write notes, prescreen patient intakes, etc. Sounds great right? Well, not exactly. That's because what YC and other investors are getting wrong about this generation of physicians versus those of the past (who did nothing to stop consolidation or electronic health records), is that they aren't just going to sit around and wait for someone else to build something for them anymore. They're builders who want to solve their own problems, and they're willing and eager to learn. Just look at telemedicine adoption data vs AI to explain what I mean: - Prior to COVID telemedicine was a <19% - During COVID it grew to > 90% - Post COVID it went back down to <40% But AI? - In Jan 24 it was 38% - Latest survey in Oct shows it >68% - By this time next year it should be near 100% This is the fastest adoption of any tech in healthcare in history, and it's only the beginning. Every day, new models, new features, and new tools are created for non-developers to build software using AI. What makes people believe that doctors can't or won't figure out how to use zero shot/few shot prompting in environments like cursor or claude just like everyone else? So if your bet is that AI doctors will make the same purchase decisions as their non-AI counterparts of the past, you're most likely making the wrong bet. This generation of doctor are builders. #HealthcareOnLinkedIn #AIagentsForDoctors

204

73 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

AI startups are commanding massive investments, with 1 in every 3 VC dollars now going to AI. Notably, Safe Superintelligence Inc. (SSI), an early-stage startup, secured a $1B round, becoming the 9th AI company to achieve a billion-dollar round this year. Despite investor confidence and FOMO, many fledgling AI startups may not meet high expectations and could face significant challenges ahead. Will AI investments continue at this pace? #AI #VC #TechInvesting #Startups #Innovation

6

Another great share from Brian R. Spisak, PhD From the article "Concepts" are crucial when it comes to "reasoning-intensive" tasks such as coding and math, where it can be difficult to specify the required concepts in a keyword-based search query. It's even harder to determine which documents might be relevant." CONCEPTS. It's all about the concepts. I learned all about Neo4j yesterday. I was surprised to learn that "graph" Neo4j doesn't automatically create the nodes, labels, relationships, or relationship definitions that power the underlying "graph data model." 🤔 I studied BPMN and UML (unified modeling language/CONCEPT modeling) over 20 years ago in graduate school under the guidance of two of the most prominent pioneers in the informatics field, Sue Bakken and Charlie Mead, MD. At that time, I didn't fully understand the significance of what I was learning. However, someone once said, "Once you learn how to abstract data into concepts and relationships to create (graph) data models, you will never see the world the same again." It's true; my brain was rewired during those two years to see the entire world as concepts/graph data models. My daughter gets annoyed when I complain under my breath about poorly designed airport processes based on flawed underlying graph data models. I've been viewing the healthcare industry, its functions, and the "data" as graph models for over 20 years. Graph data modeling is all about "metadata." Everyone uses the word metadata without understanding what it means. It's about creating the context of the data, which is often stripped away and discarded during our customary ETL processes and current relational database designs. Context enriches the data by adding back the meaning of the data when originally collected, including the who, what, when, where, how, and why of the workflow - the most atomic unit of healthcare activity. Context is necessary to determine which data is suitable for analytics, digitizing clinical practice guidelines, clinical decision support, quality measurement, AI training data, ML, GenAI, and creating the blueprint for aggregating and harmonizing disparate data sources. Creating a graph data model requires domain expertise; it's not the work of computer scientists or software engineers. It's crucial to understand how this new technology works, what it does, and what it doesn't do. If you need help with concept/graph data modeling in healthcare, feel free to send me a direct message Finally, as my dad used to say, "Don't let anyone give you a wooden nickel."😅 #HealthAI #CDS #CPG #eCQM #neo4j #BPMN+ #UML #OGDataHarmonizationQueen #AppliedInformatics

4

4 Comments

Someone broke down the estimated costs of a $15 Sweetgreen salad. We need these cost graphics for digital health to help future founders. Especially for tech enabled cos, the greatest cost is labor. Much of the "value created" is in curation and matching, which means the gross margins are reliant on labor arbitrage. Pay someone $X, but get paid $X+Y% per visit. The challenge with healthcare is that you often have to pick one of two business models: 1. specialized, low volume, high cost, low gross margin % 2. generalized, high volume, low cost, high gross margin % I'd argue founders should focus on gross dollars over the lifetime of the customer (like Jeff Bezos said) instead of gross margin %'s, but we'll leave that for another time. The challenge is that many tech enabled digital health cos have high cost of goods sold (labor) and moderately high product, design, and engineering budgets. We must build differentiated solutions for patients, clinicians, and clients to be out incumbents, however, we're realizing that companies are ultimately valued on the discounted value of future cash flows. And the high SG&A costs at most organizations might be inbalanced to the unit economics of the business. I'm not sure the math pencils out for everyone unfortunately. But I'm just a guy on the internet, would love to hear the thoughts from those smarter than I (you!). Comment below. --- p.s. I have no idea the accuracy of the graphic and not an investor in Sweetgreen. Cheers. Credit: David Crowther

112

27 Comments

On the latest Vital Signs, Nikhil Krishnan and I had a ton of fun chatting with Dr. Atul Butte. Dr. Butte is a professor at UCSF, Chief Data Scientist of the UC Health System, and a co-founder of numerous healthcare startups. In our conversation, Dr. Butte shared his takes on healthcare AI opportunities, challenges, hype, and more. A few highlights: ⚡LLM use cases, data, and growth LLMs are already helping to solve the “drudgery” aspects of medicine, such as automatically drafting SOAP notes and patient messages. Dr. Butte is excited about future applications of LLMs that can help with the “cognitive” aspects: with difficult medical problems an LLM might be able to stitch together disparate pieces of information to generate a potentially novel solution. He also highlights that prompt engineering will become the next “coding language” as so much of leveraging LLMs is knowing how to ask them the right questions. Additionally, Dr. Butte discusses how since 2016, UCSF has been working to deidentify its patient notes so that it can query them to answer clinical questions and ultimately enhance patient care. ⚡The need for patient-facing AI As patients are increasingly able to access their own personal health data in nicely structured formats (e.g., EHRs, wearables), they desperately need more tools to help them interpret and make decisions using that data. Dr. Butte understands that these patient-facing solutions are challenging to get funded but he emphasizes that patients need to receive the benefits of AI sooner. ⚡Evaluating healthcare AI solutions In order to evaluate AI solutions, we often put human specialists – such as medical professionals – on the same tasks. Understanding that specialists often disagree with each other, a significant challenge arises in figuring out how we can evaluate AI considering we can’t just “average” all the specialist opinions. Dr. Butte points out that there still remains a massive gap in AI evaluation tools. Additionally, there are so many organizations that want to have a say in how healthcare AI is regulated that the landscape is uncertain. An awesome conversation with an expert on all things AI meets healthcare. Listen to the full episode below: Spotify: https://spoti.fi/3Wr3643 Apple: https://apple.co/4acwgan

82

8 Comments

🧩 One thing we've noticed with #VentureIQ technology is how well it can distill and communicate complex technical material 🧩 Whether it's a #deeptech venture, a novel #AI application, or even a patent submission, we can quickly distill the business and strategic implications for investors...and sometimes even the founders/PIs themselves! 🤯 Based on our work so far, the tech can handle a wide array of research areas and technological sectors -- anything from blockchain algorithms to vaccine development. It handles the nuance of the field, the specific terminology, and extracts the key concepts which matter for venture-building. 🤓 To demonstrate, I took a patent I helped write with Eric Laber on reinforcement learning. I ran it through #Ventureiq to build a commercial assessment (key differentiators, target markets). Then for fun I had #NotebookLM make a podcast out of it. I've uploaded the result it here (DM me if you'd like the actual report): https://lnkd.in/eiTejsny This is more than just automation or streamlining operations - it's about unlocking clarity and insight when it was previously unattainable, and the ability to do it over and over again no matter the sector or vertical. Interested in more info? DM me so we can chat.

11

2 Comments

5 intriguing AI developments I learned about last week: 1: Komodo Health + LangChain It’s one thing to see AI Agents and Compound AI Systems being launched in Retail etc but seeing Komodo Health build an AI assistant that “leverages the power of NLP and GenAI to provide personalized and informative responses to healthcare queries” within a highly regulated industry is significant. Big shoutout to the rockstars over at LangChain including Harrison Chase + Jacob Lee. No surprise they're involved. Link: https://lnkd.in/gjPMcH2S 2: AbbVie Gen AI enabled Enterprise Product Incubator for Capabilities (EPIC) Same as #1. This is happening in a highly regulated industry… “What does it look like to scale Generative AI to the enterprise from a grass roots level?” “Seventy-plus of the power users and builders leveraging our AbbVie Intelligence platform capabilities met to dig into the new advances in our vendor agnostic, enterprise supported, GxP ready, secure LLM powered ILIAD platform. Congratulations to Jon Stevens and the rest of the Enterprise Product Incubator for Capabilities (EPIC) team for coordinating such a great event!” - Brian Martin Chief AI Product Owner, ACOS Senior Research Fellow at AbbVie Link: https://lnkd.in/gS2RC-Q5 3: Google's NotebookLM A quote like this most definitely gets my attention - “The best software ever created for leveraging AI in your creative work.” - Tiago Forte, author of “Building a Second Brain.” Link: https://notebooklm.google/ 4: Arcade Company “Arcade enables anyone to design, purchase, and sell custom, manufacturable products with a simple text or image prompt.” “Arcade brings together the power of generative AI with a global network of top artisans to turn user ideas into personalized, physical products in a quantity of one. Arcade aims to redefine commerce by offering unprecedented personal choice, expression, and meaning in product creation. Link: https://lnkd.in/gGud-jHD 5: “Every white collar job will have an AI co-pilot. Then an AI agent.” - Angela Strange, Andreessen Horowitz “Which roles might change most? James da Costa and I took a look at the Bureau of Labor statistics to inspire some ideas.” Link: https://lnkd.in/gMJ8VASz Preview of next week? Ethan Mollick comments below about LLM enabled ad models and the reactions are likely to make it... ;) “No one has figured out how you integrate advertising with LLM replies.” “This is sort of a big deal, given that where the sloshing pool of ad dollars flows determines almost everything about the digital spaces in which we operate.” Enjoy your Monday!

3

1 Comment

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

GenAI is rapidly transforming industries. *Blink* and you’ll miss the next development. Having elderly and remote family members, I closely observe GenAI's impact on healthcare. Healthcare has long explored AI and data to improve patient care. LLMs accelerated this progress exponentially. Now, AI is no longer a 'nice to have'—it's a 'cannot ignore.' AI is improving patient outcomes in many ways. Three examples: 1️⃣ Faster, cheaper drug discovery AI models swiftly sift molecular data, shortlisting promising drugs and simulating disease effects. 2️⃣ Personalized treatment plans 20 years ago, IBS patients received nearly identical management plans. Now, AI combines patient data, medical histories, and latest research to generate tailored recommendations. 3️⃣ Enhanced medical imaging AI assists in detecting subtle patterns and anomalies, aiding diagnosis and suggesting further investigation. Beyond these, I see three crucial categories where AI is revolutionizing senior care specifically: → AI-powered vertical software for senior living → Monitoring and prediction for aging in place → Assistive technology within senior healthcare In these areas, AI can be present wherever a human can't, significantly improving seniors' safety and well-being. Medical professionals will always make the final call. But AI is becoming an invaluable healthcare assistant—supporting doctors in making the best choices for patients. GenAI is revolutionizing healthcare across the board, from drug discovery to personalized care for our aging population. 👉 Building an AI-native company in healthcare, especially senior care tech? I'm always eager to connect with technical founders tackling this $30B market opportunity. Leave a comment or send me a message.

29

2 Comments

I am very excited to share the pre-print study on Atropos Health ChatRWD beta results(link to paper below). ChatRWD beta outperformed ChatGPT 3.5, Google Gemini 1.5 Pro. and Anthropic Claude 3 on answers to clinical questions when reviewed by independent physicians. OpenEvidence also outperformed the Big Tech LLMs by a wide margin. Most important to us, ChatRWD performed highest on Physician trust and Best Answer and was able to answer questions asked 94% of the time. As the Gen AI cycle continues to move at a rapid pace we firmly believe that "Trust" and "Accuracy" are going to be critical, and that publishing your performance on both is table stakes. It is also clear to me that in the quest to be able to answer 100% of questions with Clinical grade accuracy training LLMs solely on the literature will not be enough. We need to produce new evidence in the form of high-quality transparent studies using appropriate methodology. We are adding beta users post this announcement, so please sign up to try it, and we will have more to come on full launch this summer. Exciting times and hope everyone has a great July 4th. https://lnkd.in/es5gWK84 #generativeAI #LLM #Realworldevidence

225

13 Comments

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

🚀 Accelerating the Future of Healthcare with AI 🚀 I recently had the pleasure of tuning into an insightful episode of The Chris Voss Show featuring Lana Feng, Ph.D., CEO and co-founder of Huma.AI. 🗝 Here are some key takeaways that highlight the transformative power of AI in drug development and precision medicine: 🧠 Introduction to Huma.AI: Dr. Feng’s company, Huma.AI, leverages generative AI to automate data analysis, speeding up the development of life-saving drugs. Their mission? To reduce the time and cost involved in bringing these drugs to market. 🌟 💡 Challenges in Drug Development: Traditional drug development is a 10-year, $2.6 billion process plagued by data bottlenecks and regulatory complexities. AI can revolutionize this by handling vast amounts of unstructured data from various sources efficiently. 📊 🤖 The Role of AI: Utilizing advanced AI models like GPT-4, Huma.AI ensures data privacy, accuracy, and transparency. This leads to more reliable insights, faster clinical trial enrollments, and ultimately, better patient outcomes. 🏥 🌍 Real-world Applications: From identifying the right treatments in precision medicine to accelerating clinical trials, AI’s impact on healthcare is profound. It’s especially crucial in areas like cancer treatment and chronic disease management. 🎯 📈 Industry Recognition: Huma.AI has garnered significant recognition, being named a leader in generative AI for life sciences by Gartner. 🏆 🔮 Future Directions: The integration of AI in healthcare requires responsible usage, industry-wide collaboration, and continuous efforts to mitigate risks like data privacy issues and AI hallucination. 🤝 🌟 AI for Good: Dr. Feng emphasizes the importance of using AI responsibly to lower drug prices and improve healthcare outcomes. This approach ensures AI serves as a powerful ally in our quest for better health. ❤️ I’m excited about the potential of AI in transforming healthcare and making life-saving treatments more accessible and affordable. Let's embrace this technology for a healthier future! 🚀 #HealthcareInnovation #AI #PrecisionMedicine #DrugDevelopment #TechForGood #FutureOfHealthcare #ChrisVossShow #HumaAI #Leadership

11

After a multi-year period when it seems that venture capital funds would only get larger, CRV is moving in the opposite direction. Instead of raising a $500 million growth fund and then a $1 billion vehicle to follow it, the firm is returning just over half of its 2022-era growth fund. Why release $275 million worth of LP commits? "Market conditions have changed for the worse. And valuations for startups are too high relative to their potential payoff." Will other firms follow CRV? If exits do not recover in the next few quarters, the answer may be yes. The nuances of capital calls can play a larger role in eventual IRR than you might have thought.👇

50

5 Comments

💥Google's new Med-Gemini for healthcare outperforming GPT-4 Med-Gemini is a family of multimodal AI models built upon their Gemini models, specifically designed for the healthcare industry. From what I've read, these models are built on Google's existing Gemini AI architecture, but they've been specifically designed and trained for medical and clinical applications. It can process long-form medical records and research papers, and it can output text, images and videos to help explain things. Google is claiming that Med-Gemini has even outperformed OpenAI's powerful GPT-4 model on certain medical reasoning tasks, which is pretty impressive. On standard medical benchmarks like the MedQA exam, the Med-Gemini-L 1.0 model has scored really high, beating out even Google's own previous Med-PaLM 2 model. One cool feature is that Med-Gemini integrates web search, so it can pull in reliable information from the internet to provide more nuanced and factual responses. The models aren't publicly available yet - Google is still working on improving them and making sure they adhere to responsible AI principles around privacy and fairness before they release them. But from what I've seen, Med-Gemini is shaping up to be a really powerful set of medical AI tools.

1

New release! John Snow Labs #Healthcare #NLP & #LLM 5.5 includes 6 new Medical Large Language Models - tuned for different tasks (Q&A, RAG, NER, Summarization) and sizes. The small model outperforms GPT-4o almost 2-to-1, on medical text summarization in a blind evaluation by practicing medical doctors. Release notes: https://lnkd.in/gsiGF26i Conference talk about the blind evaluation and all results: https://lnkd.in/gx8FVuVU #ai #generativeai #datascience #healthcareai #healthai #nlproc #textmining #clinicaldata #medicaldata #responsibleai

32

1 Comment

mdhub (YC S24) is a clinical AI assistant for mental health clinicians, helping them run their clinics more efficiently. Mental health issues are a growing concern in America, with approximately 50 million Americans currently grappling with a mental health condition— a number increasing by 1.5 million every year. At the same time, the US is experiencing a massive shortage of 30,000 mental health clinicians in 2024, a situation costing the US health system $300 billion annually due to delayed treatment. Yet, clinicians are spending only 50% of their day with patients, while the rest of their time is consumed by admin-heavy work, delaying access to mental health care even further. With mdhub, mental health clinicians become 10x more efficient by automating tasks such as clinical documentation, drafting patient after-visit reports, pre-charting, and treatment planning. Over the last 6 months, mdhub has already successfully saved clinicians more than 100,000 hours in admin tasks. Congrats on the launch, Efrén A. Lamolda and Dominik Middelmann! 🚀 https://lnkd.in/gPCTzfy9

90

5 Comments