About
Activity
6K followers
-
Matt Turk shared thisMatt Turk shared thisLast week we were honored to host researchers from Google DeepMind, NVIDIA, Niantic, and Waymo to discuss issues at the cutting edge of vision and robotics. A huge thanks to Mia Polansky from our DataLab team, who helped lead a discussion that ranged from 3D scene reconstruction to the ways sim-to-real dynamics are shifting as models rapidly improve. Overcoming barriers to performance and versatility in robotics takes smarter collaboration as well as smarter data. We're pulling together additional dinners in SF, NYC and London and welcome recommendations for senior researchers in vision & robotics who could add to the conversation. #AI #Robotics #ComputerVision #WorldModels #ResearchCommunity
-
Matt Turk shared thisMatt Turk shared thisLast week we were honored to host researchers from Google DeepMind, NVIDIA, Niantic, and Waymo to discuss issues at the cutting edge of vision and robotics. A huge thanks to Mia Polansky from our DataLab team, who helped lead a discussion that ranged from 3D scene reconstruction to the ways sim-to-real dynamics are shifting as models rapidly improve. Overcoming barriers to performance and versatility in robotics takes smarter collaboration as well as smarter data. We're pulling together additional dinners in SF, NYC and London and welcome recommendations for senior researchers in vision & robotics who could add to the conversation. #AI #Robotics #ComputerVision #WorldModels #ResearchCommunity
-
Matt Turk shared thisMatt Turk shared thisWhen it comes to field marketing and events, when the team just "gets it" - it feels like magic. Waitlists fill up. Conversations flow freely. You hit fire code occupancy limits. The energy is different. True field marketers know: This doesn't happen overnight. The small dinner leads to a happy hour, the happy hour paves the way to a flagship event. Every magic-making moment creates momentum for more marketing madness. Most companies hire only after feeling the pain of *not* having the magic. Luckily, at Protege - we have the opposite problem. Too much magic - and not enough help for Melissa Lang and crew on the ground. We've run 15 activations across 6 cities the past 3 months alone. So now, we're looking for the right Field Marketing Manager based in NYC or London that intuitively understands that: 🍽️ The right rooms create lasting relationships 👹 Events are more than logistics... but the devil's in the details 👻 The perfect events are the ones where no one notices what went wrong because you handled it ⏭️ What happens after the event matters just as much as the event Link to apply in comments... know the perfect person for this role? Tag them below!
-
Matt Turk shared thisI highly recommend working with Caleb Lee - you will learn a lot! For anyone in my network interested in this role, please reach out!Matt Turk shared thisLooking for a founding technical "superPMM" to join the Marketing team at Protege! Protege is an absolute 🚀. From founding in mid-2024, to 9-figure revenue in 2026 — and it's only August! I'm looking for a π-shaped marketer who spikes in Product & Content. A scrappy, quick-thinking learner who loves writing. A superPMM who will join our rocket ship and quickly lead our product marketing from end-to-end. Someone who's has the unique combo of being kind — and being a relentless problem-solver. As Anna Furmanov puts it, a superPMM who can think strategically across positioning, messaging, and GTM...and then execute across the company themselves, using AI and tolling to leverage the heavy lifting (come build more claude skills with me!). This sound like you? Would love to chat! This role is remote-first, with competitive comp for seasoned PMMs who love to bet on themselves! Shoutout to Anna Furmanov for the superPMM term, and Emily Kramer for the π-shaped marketer kramework! https://lnkd.in/gpCryZKU
-
Matt Turk reposted thisMatt Turk reposted thisAt Protege, we’re seeing what many others in healthcare are seeing: There’s a huge clinical readiness gap - the gap between how health AI performs in test settings and how it performs in clinical workflows. Leading academic institutions and health systems alike all face a similar issue: How do we best measure AI for healthcare applications when AI is already being used by doctors and patients alike? This isn’t an easy question. But it’s an important one that stands between AI and real-world deployment at the point of care. On some healthcare tasks, AI already performs as well as the best clinicians in the world. On others, the model struggles to recall a simple lab result from a simple report. Protege Co-Founder and Chief Scientific Officer Engy Ziedan wrote about this for The AI Collective in their most recent newsletter — in it, Engy explains how she and the rest of Protege’s DataLab team approach this challenge from a few different perspective: #1 - Test Design: Does the test measure the “right” answer? #2 - Data: Does the data represent real-world scenarios? #3 - The Referee: Who decides which model is best? #4 - Gamification: What happens when passing the test becomes the goal? As the industry moves beyond static test questions and toward realistic, multi-turn scenarios built on real-world healthcare data, we increasingly need an independent evaluator of how AI models are performing. Check out Engy’s full write-up with The AI Collective here: https://lnkd.in/gkx9zeBD
-
Matt Turk shared thisExcited to share this piece that our DataLab team at Protege published with Andreessen Horowitz to understand how to build healthcare AI benchmarks that reliably produce models that produce real health gains, not just good test scores. Outside of difficulties in creating high quality benchmarks, scoring 100% on benchmarks != being a good doctor in the real world (which is "the verification ceiling" as Bobby Samuels aptly puts it) Clinical decisions are full of subjective “taste,” recency bias, and idiosyncratic physician styles. Static tests can’t capture long-term patient outcomes. As the piece illustrates, the data on surgeon variation in knee replacements, prompt sensitivity flipping leaderboards, and how fast patients + note-writing are already adopting AI illustrate the urgency of getting this right.Matt Turk shared thisAI in healthcare is here. More than 300 million people use ChatGPT for healthcare questions every week. Protege DataLab analysis shows that nearly a third of all SOAP notes are now written by AI. Right now, AI is now extraordinarily good at tasks where you can train it on “the right answer.” But what if there isn’t one “right” answer? Are the AI models of today actually good doctors… or are they just good at the test? This is the challenge in healthcare AI today: the verification ceiling. Benchmarks have hit the issue medicine has noticed decades ago: trials measure efficacy, but the real world cares about effectiveness. Today, benchmarks lack even a good measure of efficacy and are far away from effectiveness. Training data and compute may not be the only binding constraints. Instead, we are constrained by learning how to know when the model was right. Protege’s DataLab dug into the data with the team at Andreessen Horowitz to understand how to build healthcare AI benchmarks that reliably produce models that produce real health gains, not just good test scores. Read the full analysis here: https://lnkd.in/gYJAamZ8 Engy Ziedan Protege Allison Fox Matt Turk Zoe Li Alex Muir Caleb Lee
-
Matt Turk reposted thisMatt Turk reposted thisWe brought ~100 operators and researchers in Frontier AI together at A16Z to discuss the State of Training Data. The conversation represented a range of perspectives, from researchers at foundation labs like OpenAI, NVIDIA, Microsoft and Meta to applied AI operators at Cognition, Databricks, PayPal and WHOP to reinforcement learning gyms such as HUD, SF Data, and AfterQuery, as well as Protege's own real-world data partners. Very grateful for the group that gathered last week. The team has learned that the field needs more events to bring the training data ecosystem together! Photos below - follow team accounts like Engy Ziedan, Logan Johnston, Allison Fox, Richard Ho , Bobby Samuels, Dave Davis and Matt Turk for content from these events and their insights on AI x Training Data.
-
Matt Turk shared thisMatt Turk shared thisWe brought ~100 operators and researchers in Frontier AI together at A16Z to discuss the State of Training Data. The conversation represented a range of perspectives, from researchers at foundation labs like OpenAI, NVIDIA, Microsoft and Meta to applied AI operators at Cognition, Databricks, PayPal and WHOP to reinforcement learning gyms such as HUD, SF Data, and AfterQuery, as well as Protege's own real-world data partners. Very grateful for the group that gathered last week. The team has learned that the field needs more events to bring the training data ecosystem together! Photos below - follow team accounts like Engy Ziedan, Logan Johnston, Allison Fox, Richard Ho , Bobby Samuels, Dave Davis and Matt Turk for content from these events and their insights on AI x Training Data.
-
Matt Turk reposted thisMatt Turk reposted thisWhat happens when AI becomes too smart to evaluate? From Engy Ziedan's keynote talk at The State of Training Data at A16Z.
-
Matt Turk liked thisMatt Turk liked thisAfter the best 9.5 years of my professional life, and with a heart bursting with gratitude, I'm spinning out of FJ Labs to start my own firm. There is no place in the world like FJ. Fabrice Grinda and Jose Marin helped pioneer super-angel investing, a collaborative, network-driven model that still sounds unfathomable when I describe it. I had the privilege of working on almost 500 investments and raising capital from LPs in 20+ countries. For the rest of my life I'll remember the late-night debates, the thesis-building (and debauchery) at offsites, and working with a group of colleagues who became extremely close friends. Fabrice and Jose - thank you for taking a chance on a junior IR associate and trusting me with your firm. From hunting in Spain to kitesurfing at Fabrice's villa in Turks and Caicos, fundraising from Riyadh to Bangkok, raising a SPAC (oops!), and Halloween in Shanghai: what a ride. Arne H. Halleraker, my Nordic brother from another mother - you were the best partner I could ask for. You bring an engineer's rigor to investing, and taught me so much about management, science, technology and strange Norwegian traditions :) I'm convinced you can go down as an all-time great frontier tech investor. The team is in the best hands possible. Matthew S., the title COO doesn't do you justice. Ariel Lebowits, I hope to never negotiate against you :) William in Guillouard, the funny Frenchman and king of IT. Lauren Lee, Danny Browne and Mehul K., the young guns who keep me honest on AI. Ryan Geary, Owen McCarthy, CPA, Morgan Spar, Gloria Rodriguez- the best ops team in the game. My FJ teammates, both former and current- thank you for everything. I consider you close personal friends and always will. To my founder and investor friends (too many to list here, but you know who you are!!) - while my day-to-day at FJ is ending, I'm staying in the family as a Senior Advisor (think venture partner on steroids), helping with the portfolio, and will work closely with the FJ Labs team on secondary opportunities. As for what's next- if you know me, you know I'm an obsessive distance runner. An ekiden (eh-kee-den) is a long-distance relay race that started in Japan more than a century ago. Each runner carries a sash called the tasuki for one leg, then hands it to the next teammate. As I wrote in Venture's Nifty Fifty Moment, great companies now stay private longer than the funds that backed them were even built to last. Venture has quietly turned into a race with no finish line. I like to say that one of the polite lies of venture capital is the 10-year fund life. Plenty of early investors, employees and LPs have run their leg, and they're still waiting for someone to take the sash. Announcing Ekiden Capital. If you're looking for the next leg in the venture relay race, my inbox is open.
-
Matt Turk liked thisMatt Turk liked thisScams are through the roof, thanks to AI. Never trust an unverified LinkedIn profile. Only partially trust a verified one. If they just joined and have few connections, that's a big red flag. We receive dozens of "business" decks looking for money each week. Sometimes we'll take a look just to see what kind of scam they are running. In one case, the CEO didn't realize that most of their "revenue" was coming from North Macedonian accounts. Multiple dumb VCs invested in them. LinkedIn needs to do a better job of screening. The Federal Bureau of Investigation (FBI) needs to do a better job of putting them in prison. Pangram Labs detects AI as the first line of defense.
-
Matt Turk liked thisClaude Marketplace just expanded into a single home for all the best of the Claude ecosystem! Claude is powerful out of the box, but a lot of real problems need the right tools, the right data & the right partners to help solve them. Now the marketplace includes 2,000+ connectors & plugins, leading Claude-powered products (bought with committed Anthropic spend) and service partners. Explore the ecosystem: claude.com/marketplaceMatt Turk liked thisYou can now discover tools, agents and expert partners on Claude Marketplace. Use it to: - Connect the tools your team already works in, like Figma, Notion, and Slack. - Buy products built on Claude from companies like Cursor, CrowdStrike, and Snowflake. - Or find a service partner, like Accenture, Boston Consulting Group, or Deloitte, to help you roll Claude out across your company. Browse the marketplace: claude.com/marketplace And if you're building for Claude, list your tools, agents or services here: https://lnkd.in/eRj-k48M
-
Matt Turk liked thisMatt Turk liked thisExcited to share that I will be speaking at the LA Games Conference on September 30 in Westwood! (Link to event in comments.) I'll be joining a panel titled "Licensing in the Age of AI: Legal Frameworks for Modern Game Development." This conversation will dig into a variety of areas, including the relevance of game-related data as training data for AI. My co-panelists will be: John Kulback, Loeb & Loeb, Jesse B. Levin, Glaser Weil, Joanna Popper, media and tech executive, and moderated by Cooper Jackson from Electronic Arts (EA). (John, Cooper and I were all colleagues at Fox.) If you're attending LA Games Conference and interested in exploring the world of AI training data, I would love to connect. Grant Murphy-Herndon Rahman Malik Brian J. Cappy Xenia Shevnina June Yeoh Carole Pigeard
-
Matt Turk liked thisMatt Turk liked thisLast week, I got to discuss the future of AI in healthcare with Eva Steinman Ziad Sultan Ray Wang. Conversations like this are a good reminder that while the industry has moved incredibly quickly these past few years, we still have a lot to accomplish. Some of the most exciting innovation is happening at Protege and if you want to get involved, we're hiring across the board! https://lnkd.in/gj4wU4KV
-
Matt Turk liked thisMatt Turk liked thisLast week, I got to discuss the future of AI in healthcare with Eva Steinman Ziad Sultan Ray Wang. Conversations like this are a good reminder that while the industry has moved incredibly quickly these past few years, we still have a lot to accomplish. Some of the most exciting innovation is happening at Protege and if you want to get involved, we're hiring across the board! https://lnkd.in/gj4wU4KV
-
Matt Turk liked thisMatt Turk liked this"Who is making sure that AI answers in a doctor's clinical AI tool is safe and correct?" Frontier models care - but ChatGPT or Claude shouldn't grade itself when it's deployed in a health system Specialized healthcare AI model builders care - but they have the same grading concern Healthcare system and providers care - but they're actively looking for the source of truth for whether models are safe for their context and environment Academics care - and continue to debate health AI effectiveness while looking for representative real-world data The US government is starting to care - HHS just appointed a deputy commissioner for technology and artificial intelligence So everyone cares. But there hasn't been an incentives-aligned, expert healthcare organization that exists to help ensure AI alignment that is also able to help improve the models at their training source. This is something Engy Ziedan and Protege's DataLab researchers think about daily. With Protege's healthcare data, expertise, and scale, we're actively working with model builders to help solve this pressing issue. Explanation and interview with Engy, Daisy Wolf, Eva Steinman on the a16z Bio + Health Raising Health podcast linked below!
Experience & Education
Licenses & Certifications
Volunteer Experience
Publications
-
Counterfactual Evaluation Reveals Hidden Capability Profiles in Clinical LLMs and Agents
RLEval Workshop @ ACM Conference on AI and Agentic Systems (CAIS) 2026
See publicationAccepted to the inaugural RLEval Workshop at ACM CAIS 2026 (Conference on AI and Agentic Systems) and selected for an invited talk based on top ratings from reviewers.
The paper introduces the **Causal Sensitivity Score (CSS)**, a pre-registered interventional metric that mutates oncology tumor-board cases along five clinically meaningful dimensions and measures whether a model updates its recommendations in the pre-registered correct direction.
Benchmarked against the published…Accepted to the inaugural RLEval Workshop at ACM CAIS 2026 (Conference on AI and Agentic Systems) and selected for an invited talk based on top ratings from reviewers.
The paper introduces the **Causal Sensitivity Score (CSS)**, a pre-registered interventional metric that mutates oncology tumor-board cases along five clinically meaningful dimensions and measures whether a model updates its recommendations in the pre-registered correct direction.
Benchmarked against the published coverage-based **Consensus Match Score (CMS)** across 224 tumor-board cases, CSS produces nearly the opposite ranking of six frontier models from three labs. The evaluation also reveals a universal safety blind spot: every frontier model performs poorly on surgery-status interventions (≤17.2% CSS), a failure mode that coverage-based evaluation does not detect.
The metric also transfers to tool-using agents. Cross-judge replication and three-rater medical-professional validation confirm the aggregate findings.
Patents
Courses
-
AP Biology
-
-
AP Calculus AB
-
-
AP Calculus BC
-
-
AP Computer Science
-
-
AP English Language and Composition
-
-
AP English Literature and Composition
-
-
AP Environmental Science
-
-
AP Human Geography
-
-
AP Macroeconomics
-
-
AP Physics B
-
-
AP Psychology
-
-
AP Spanish
-
-
AP Statistics
-
-
AP US Government
-
-
AP US History
-
-
AP World History
-
-
CS160 - User Interfaces and Design
-
-
CS170 - Efficient Algorithms and Intractable Problems
-
-
CS186 - Introduction to Database Systems
-
-
CS188 - Introduction to Artificial Intelligence
-
-
CS189 - Introduction to Machine Learning
-
-
CS195 - The Social Implications of Computing
-
-
CS399 - Professional Preparation: Supervised Teaching of Computer Science
-
-
CS61A - The Structure and Interpretation of Computer Programs
-
-
CS61B - Data Structures
-
-
CS61C - Great Ideas in Computer Architecture
-
-
CS70 - Discrete Math and Probability Theory
-
-
Develop with Ruby on Rails
-
-
EE16A - Designing Information Devices and Systems I
-
-
EE16B - Designing Information Devices and Systems II
-
-
Front End Foundations
-
-
IEOR 190A - Advanced Methods of Berkeley Entrepreneurship
-
-
IEOR 191 - Technology Entrepreneurship
-
-
IEOR 195 - Newton Lecture Series
-
-
Learn Rails APIs
-
-
Learn iOS Development
-
-
Master Objective-C
-
-
Math 54 - Linear Algebra and Differential Equations
-
-
Math53 - Multivariable Calculus
-
-
Physics7A - Physics for Scientists and Engineers: Mechanics
-
-
Physics7B - Physics for Scientists and Engineers: Thermodynamics and Electromagnetism
-
-
Sociology 111AC - Sociology of the Family
-
-
Sociology 3AC - Principles of Sociology: American Cultures
-
-
Theater R1B - Performance Writing and Research
-
-
iOS Programming
-
Projects
-
FitPay (Money 20/20 Hackathon)
We combined FitBit’s live data collection with Vantiv’s seamless payment technology to execute payments in real time. Our technology is used to further engage athletes, increase competition participation rates, related charitable donations, and transaction numbers; thereby creating value for all stakeholders—athletes, organized competitions, charitable organizations, FitBit, and Vantiv
Other creatorsSee project -
SmartScope (iPhone Application) - Winner of most creative hack!
There are not many products that allow blind people to perceive the world around them. For them, their lives are ruled by the sounds they hear, while for many of us, it is what we see. That's why we brought SmartScope. The idea behind SmartScope is that a blind person (or anyone who wishes to use the application), will turn on the application. A camera will take a photo every 5 seconds or so, and will send the information to IBM Bluemix. Using their Visual Analytics Watson, we convert this…
There are not many products that allow blind people to perceive the world around them. For them, their lives are ruled by the sounds they hear, while for many of us, it is what we see. That's why we brought SmartScope. The idea behind SmartScope is that a blind person (or anyone who wishes to use the application), will turn on the application. A camera will take a photo every 5 seconds or so, and will send the information to IBM Bluemix. Using their Visual Analytics Watson, we convert this image into a parsable JSON. We then take this JSON, and put that through another Watson, the Text to Speech, which converts the JSON into speech. Through this, anyone who uses the application will perceive the world around them through simply auditory statements.
Other creators -
Snip!t
Visually Appealing Consolidated Media Reader.
Pulls RRS Feeds, planning to add Social Media Integration
Implementation: JavaScript, CSS, HTML5, Flask
We wanted to be able to create a home page for someone who just wanted to be able to catch up on all their important news and information without having to go tab to tab but still give them as much functionality as possible. With that we created Snip!t, where a user could keep track of all their favorite rss news feeds on a crisp…Visually Appealing Consolidated Media Reader.
Pulls RRS Feeds, planning to add Social Media Integration
Implementation: JavaScript, CSS, HTML5, Flask
We wanted to be able to create a home page for someone who just wanted to be able to catch up on all their important news and information without having to go tab to tab but still give them as much functionality as possible. With that we created Snip!t, where a user could keep track of all their favorite rss news feeds on a crisp, clean and intuitive UI.Other creatorsSee project -
Motion Sensor Controller
We created a motion detecting safety system using an Intel Edison computer, and an IR sensor. The sensor detects motion within a certain distance and sends a text message to a desired phone with a set message notifying the user; for example, a typical message would be that the sensor has detected movement. We think this project and idea has potential to be detect home intrusions or other unsafe activities.
Other creators -
Motion Sensor Controller
We created a motion detecting safety system using an Intel Edison computer, and an IR sensor. The sensor detects motion within a certain distance and sends a text message to a desired phone with a set message notifying the user; for example, a typical message would be that the sensor has detected movement. We think this project and idea has potential to be detect home intrusions or other unsafe activities.
Other creators -
Motion Sensor Safety Controller
We created a motion detecting safety system using an Arduino microcontroller, Intel Edison computer, and an IR sensor. The sensor detects motion within a certain distance and sends a text message to a desired phone with a set message notifying the user; for example, a typical message would be that the sensor has detected movement. We think this project and idea has potential to be more effectively utilized to take precautions against home intrusions or other unsafe activities.
Other creators -
Terminus (iPhone application)
I made an iPhone app that helps people find jobs, internships, and volunteer services.
-
Webpage
-
Made a webpage that lets you access any of your favorite websites by clicking on picture icons (without typing in URLs or even the use of a keyboard at all) after saving all of your picture icons all across the webpage and by allowing the web browser to stay open on the web page after linking to another site.
Honors & Awards
-
HackDFW: Best Use of IBM Bluemix & Most Creative Hack
HackDFW
Won the prize for the "Best Use of IBM Bluemix" at HackDFW in Dallas, Texas. Additionally, the hack was awarded the highest rank in creativity among all 80+ submissions.
For details on the project see "HackDFW: SmartScope" in the Projects section. -
Army Athletic Scholarship Award
United States Army
Award given out by the U.S. Army to a student-athlete who upholds academic excellence and success while playing a varsity sport in high school.
-
Department Awards - Math, Science, History, English, Economics, Spanish
Barron Collier High School
Awards given to the top-scoring and highest achieving student in each field, every year. I was awarded the departmental award for many of these subjects for various courses in the years of 2011, 2012, 2013, and 2014.
-
Student of the Year
Barron Collier High School
Award given to the student who best represents academic excellence, integrity, success and honor in the high school class.
-
Double Varsity Basketball Letter Award
Barron Collier High School
Award given to a varsity basketball player who has earned two varsity letters throughout the course of high school.
-
1st Place in 2014 Collier County Quiz Bowl Competition
-
Award given to the first place team in the 2014 Collier County High School Quiz Bowl competition.
-
1st Place in Collier County Team Placement at High School Regional Math Competition - Calculus
Florida Gulf Coast University
Award recognized for 1st place for teams from Collier County in this high school regional math competition for the calculus division.
-
Harvard Book Award
Harvard Alumni Association
Award given to top-performing students in 11th grade who “display excellence in scholarship and high character, combined with achievements in other fields."
-
JFK Profiles in Courage National Essay Contest Finalist
John F. Kennedy Presidential Library and Museum
Distinction given to those who were in the top 1% in standing for best writing an essay that describes and analyzes an act of political courage by a United States elected official that includes analysis of the obstacles, risks, and consequences associated with the act and can be at the local, state, national, or international level.
-
Coaches' Award in basketball
Barron Collier High School
Award given to a student-athlete who best demonstrated perseverance, commitment, and hard work throughout the year in a high school sport.
Languages
-
English
Native or bilingual proficiency
-
Spanish
Limited working proficiency
Recommendations received
3 people have recommended Matt
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content