If you like to circus, buzz me about our studio 1D1 in Brooklyn.
- (Aug 2023) Follow up to communication with a character. Now with hypotheses for questions 3 (”lead, not just respond”), 11 (”who am I talking to?”), 12 (”what I feel comes out in my voice”), and 13 (”what I feel comes out in my face”).
- (June 2023) Bring to life the world around you. A mobile app + character creation tool for turning local buildings and art into conversational partners.
- (April 2023) A communicative audiovisual AI that feels just like you’re talking to a [likely cartoon] character with a human behind the curtain. I then list businesses that this could spur and what plausibly comes after this exists.
- (April 2023) Fourteen research questions towards achieving human-like Communication with a robot.
- (April 2023) Large language models are fantastic at modeling Other intent and have little need or reason to model Self intent.
- (Dec 2022) Primer is a digital AI teacher, an Aristotle, for everyone. This piece looks into how you would build it today from the perspective of a well funded startup.
- (Nov 2022) A study of Frax completed jointly with Justin Cullen and Dino Mihalopoulos. We first analyze the Frax system and then synthesize it as a piece of the larger system that is Web3 banking.
- (Aug 2022) A study of Web3 Stablecoins, with a focus on the Collateral Ratio and how it affects the resulting Behavioral Economics.
- (July 2022) A study of 0x completed jointly with Justin Cullen. We first analyze the 0x system and then synthesize it as a piece of the larger system that is Web3 Order Flow.
- (Jan 2022) A thesis motivating Web3, with the perspective that Assets are Products. Result is a clearer understanding of what is and is not worth building on blockchains.
- (Aug 2021) At the intersection of ML and Web3, explore a world where autonomous agents can find and trade with humans as equals, contextualized in autonomous vehicles, art & NFTs, and gaming.
- (Nov 2020) A machine learning write-up assessing whether we could fix a spurious correlation in a dataset by just collecting some examples and fine-tuning on those examples. We also had interest in understanding what changed as we did this and whether the correlation became spurious in another way.
- (Oct 2020) I went on a retreat and found some kinetic magic worth talking about. Lots of hallucinating, no drugs involved.
- (Sep 2020) A machine learning product idea where we reclaim the world's communal spaces by turning all the walls into talking characters at the behest of street artists.
- (Aug 2020) An abstraction of TikTok by equivocating the concept to photoshop vis a vis the layers. Assuming shared ownership of all videos and extensive tools to manipulate the content, I describe eleven different ways to collaborate on a video and what that entails / engenders.
- (July 2020) A machine learning write-up hypothesizing how to automatically yield background, character, and animation from a video.
- (April 2019) A lengthy exploration into important historical papers in machine learning meets game theory. Of particular note is a deep dive into the origins of the Fermi distribution wrt its ubiquitous use as the underlying probability of whether agent A adapts agent B’s strategy. Everyone who used this approach, including Alpha-Rank most recently, can be traced to Blume '93.
- (Dec 2018) A machine learning write-up hypothesizing how we can use humans in the loop to create unsupervised models with axes of variation about which we care greatly.
- (July 2018) A machine learning write-up hypothesizing what is the difference between amateur movement and expert movement, with a focus on handstands.
- (March 2010) A eulogy for a man who would been a best friend. Originally written for MIT's Tech Paper.
- We ran a fellowship sponsored by Jane Street - much love - where we gave grants to students around the world to run learning groups and globally share the curricula.
- Understanding research papers is hard. We tried a new way where you build the tree of concepts and then learn them going forward. We then shared the curricula and our experience with the world.
- Full compendium is on the website.
- Research into multi-agent approaches, in particular in machine learning, is overwhelmingly for 1v1 zero-sum games. Pommerman is a response where we created a team cheap-talk 2v2 game based on the famous Bomberman game.
- We ran three global competitions, including at NeurIPS 2018 and 2019.
Space of Motion
- We built a service for helping athletes with their training videos via machine learning, e.g activity recognition and video search.
- Focused on gymnastics and helping the current generation of USA olympic competitors.
- Answer questions on machine learning with chat. Was my best marketing hack yet. Thanks Intercom!
holdpls (2023 → Current) Blueprint Forest: Early Stage Investing (2022) PhD CS NYU (Machine Learning), FAIR + NVidia (2017 → 2022) Google Brain: ML Research (2016 → 2017) Machine Learning Startups + SPC (2013 → 2015) Quora (2011 → 2013) Value Investing (2010 → 2011) MIT: Math + CS (2006 → 2010)