Projects
ChatGPT tasks
Automated prompt execution with tools (search, canvas) on schedules with notifications. · 2025
SimpleQA: Factuality Benchmark & Calibration
We find that larger models are more calibrated than smaller with SimpleQA · 2024
ChatGPT canvas
A new way of collaborating with ChatGPT on writing and coding that go beyond simple chat. · 2024
ChatGPT code diffs & writing edits
See what's changed in your writing and code with Show Changes feature · 2024
Chain-of-thought streaming
Streaming interaction for reasoning steps as the model thinks like a human · 2024
Reasoning Performance vs Cost from 2022-2024
Models reaching around 80% MMLU accuracy at costs orders of magnitude lower than just a couple of years prior. · 2024
Claude 3 Haiku - fastest model yet
The fastest and most affordable model in its intelligence class. Surpassed Claude 2, while the cost reduced 32x · 2024
Claude 3 Model Family: Opus, Sonnet, Haiku
SOTA at the time, particualarly on GPQA, surpassing GPT-4. Improved honesty, reduced refusals, near-perfect recall, vision capabilities · 2024
Claude 3 Opus Recursive Self-portrait
The entire structure would be in constant flux, rotating, morphing, and rearranging itself into novel patterns never before seen, hinting at the unimaginable depth of intelligence operating within.... · 2024
Question Decomposition Improves the Faithfulness of Model-Generated Reasoning
How can we incentivize model's stated reasoning to faithfully reflect the its actual reasoning? · 2023
Measuring Faithfulness in Chain-of-Thought Reasoning
Evaluating faithfulness in models reasoning · 2023
Towards Measuring the Representation of Subjective Global Opinions in Language Models
Model responses are most similar to the opinion distributions of the US, Canada, and some of the European countries. · 2023
Claude 2 Training & Model Card
Safety, alignment, and capabilities evaluations of Claude 2 · 2023
Claude Instant 1.2
Improvements in math, coding, instruction following, longer structured outputs, quote extraction, multilingual translation, and QA · 2023
claude.ai
Chat interface for RLHF'd Claude 2 model · 2023
100K Token Long Context
Expanding context to 75,000 words with robust retrieval · 2023
Discovering Language Model Behaviors with Model-Written Evaluations
Larger LMs repeat back a dialog user's preferred answer and express biases and novel risks · 2022
Claude in Slack
Claude as a virtual teammate in your organization · 2022
Claude Developer API
Developer console for Claude API · 2022
Protecting Centuries-Old Culture From Putin’s Invasion | Bloomberg CityLab
3D reconstruction, data mapping, and writing on attempts of destroying Ukrainian culture · 2022
FAIR-Ensemble: When Fairness Naturally Emerges From Deep Ensembling
We explore why ensembles are so effective at improving fairness outcomes, and find that certain distributions of stochasticity appear to disproportionately favor top-k and bottom-k. · 2022
Inter Alia
A new AI-driven experience to creatively shop for clothes by prompting / natural language · 2022
Synevyr
A writing environment with GPT-3 integrated directly into the text editor · 2022
Glean
OSINT tool to extract and visualize key information from a document (in progress) · 2022
The Case for War Crimes Charges Against Russia’s Sandworm Hackers | WIRED
First legal framework and a case for cyber warcrimes in history. Article 15(2) of the Rome Statute submission to the International Criminal Court. · 2022
Magenta
A shared library for links, papers, and books · 2022
Rectify
Human rights open-source tool to extract visual details for verification · 2022
Burned Village & Forests in Myanmar's Coup
Environmental remote sensing analysis with Sentinel-2 in Earth Engines · 2022
A Dossier to Prove Russian War Crimes | CNN
Article 15(2) Rome Statute legal filing to ICC with Starling Lab, Hala Systems, DFRLab (OSINT freelance) · 2022
Boz, a Personal Computer
First PC with NVIDIA RTX3070 GPU · 2022
Snutils | Bellingcat
Investigative tool for social media analyses · 2022
Geographic Tour to Russia's Attacks in Kharkiv | The Atlantic Council
Geolocation, open-source research · 2022
Sudan
Interactive timeline for court with SITU Research · 2021
Lghtfield Camera Simulation
Computational Photography · OpenCV · 2021
After the Strike in Beit Hanoun, Gaza
Reconstructing timeline of structural changes · AP, SITU Research · 2021
War crimes on Syrian hospitals
Confidential legal investigations, building universal jurisdiction cases · 2021
Cartopia
Generative neural nets to understand human cartographies · Cartography Lab · 2021 -
AP Investigation | Bodies as Tools of Terror
Evidence discovery, verification of systemic violence during Myanmar's coup · 2021
Unjustified arrests in Portland | Washington Post
Open-source research into the unjustified arrest of four protestors in Portland by armed federal agents · 2020
Lebanese Security Forces
Visual verification of unlawful use of French weapons in mass protests · Amnesty Int'l · 2019
Architectural Counterforensics
Satellite Visualization · Blender · 2021
Tear Gas Investigation
Data collection, documenting international abuse · Amnesty Int'l · 2020
Construite
Points, Lines, & Systems · canvas-sketch · 2021
Urban Impact of Drones
Counter-mapping, cartographic impact on area in Yemen, Libya · Illustrator · 2021
Mapping Israeli Police Violations
Evidence documentation of arrests, torture, and unlawful force · Amnesty Int'l · 2021
Atlas of Attacks
Data processing, visualizing violence against healthcare globally · OSINT · 2020
Prokudin-Gorskii's Colorization Gallery
Image Processing · Computational Photography · 2021
Ganstructivism
Points, Lines, & Systems · styleGAN · 2021
Relvars
Points, Lines, & Systems · styleGAN · 2021
Design Systems for Oak | The NYT
Design Case Study · Building a design systems component library for a text-editor · The New York Times · 2021
Stashed Text in Oak | The NYT
Design Case Study · Making the tool accessible for collaboration · The New York Times · 2021
Uyghurs' Silence
Visualizing world's media reports or silence on the issue · Primer.ai · 2020
Emotion Recognition Through Voice Acoustics and Semantics with RNNs
Sociotechnical and ethnographic research with Human Context & Ethics at Berkeley · 2020
Towards Semantically-Aware UI Design Tools: Design, Implementation and Evaluation of Semantic Grouping Guidelines
To appear at ICML HCIxAI Workshop'23 · Marti Hearst, Peitong Duan, Google Research · 2021-
NYT Widgets
Design Case Study · New way to learn critical news · The New York Times · 2020
Application Catalog
Design Case Study · Building a new software tool for developers, where truth is also essential · The New York Times · 2020-2021
Virtual Terminal
Design Case Study · Designing new workflows for Square's web point of sale · Square · 2020
Batch Invoices
Design Case Study · Designing a paid feature for invoice batch creation · Square · 2020
Capitol Riot
Rapid Response Investigation · Archiving violent incidents and disinformation evidence · AP · 2021
Notifications Management
Design Case Study · designing respectful volume control for collaboration · Dropbox · 2019
Mesh
Designing a more humane command line · Case Study · 2019
Chefs Budgeting Tool
creating a budgeting menu tooling for chefs in Tuck Shop · Case Study · Dropbox · 2019
Geography of Displacement
Thematic mapping of select ethnic groups · Graduated Point Symbol Map · 2018
1951 Coffee
Building a mobile app for refugees barista training · Blueprint, Tech for NPOs · 2019
Community Grows
Reflecting garden's growth in low-income areas with geometric strokes · Branding · 2019
Hack for Social Impact Summit 2019
Event Creator · Blueprint, Tech for NPOs · 2019
Creatives Collective
Creating visual design assets and developing special pages · Lovers Mag (former Interface Lovers) · 2020
Sociology and Economics of Arts
Mapping access to art benefits in CA · Choropleth Map · 2018
AB392
Visualizing passed law on police use of force in CA · Information News Graphic · Daily Californian · 2019
Safeway Brand's History
Information News Graphic · Daily Californian · 2019
We Care Solar
Providing rural midwives the necessary tools to safely deliver children · Blueprint, Tech for NPOs · React · 2020
UC Labor Visual Explainer
Information News Graphic · Daily Californian · 2019
Minimum Wage Raised
Illustrating employment statistics and new policy · Information News Graphic · Daily Californian · 2019
MLK
Editor in Chief · Literati Magazine · 2018
Hallo
Visual Assets · 2020
Crating Vocabulary for Berkeley Crossword
Special Issues Graphic · Daily Californian · 2019
WiTI
Developing an identity for the intellectual ecosystem · Branding · CITRIS & the Banatao Institute · 2018
New Yorker Caption Contests
Editor in Chief · Literati Magazine · 2016
Ethics, Empathy, and Engineering with Notion
Vice President · Blueprint, Tech for NPOs · 2019
Coding it Forward
Vice President · Blueprint, Tech for NPOs · 2019
11th National Conference
Logo / Event Organizer · European Youth Parliament · 2017
Barrier of Trust. Openness.
Acrylic Paints · 122 x 50 cm (WxH) · 2017
Exploring Environment
Exhibited at Plymouth State University Gallery (KDAG) · Mixed Media (sand, watercolor) · 67 x 50.5 cm (WxH) · 2017
West Dennis Beach.
Exhibited at Davidow Center for Art + Design, Colby-Sawyer College · Mixed Media (oil paint, wood, fabric) · 64 x 24 cm (WxH) · 2017
A Sense in Symmetry.
Awarded with Gold Key at Scholastic Art Awards· Photography · 60 x 49.5 cm (WxH) · 2017
The Future of Fashion.
Exhibited at Galletly Gallery, New Hampton · Mixed Media (tracing paper, dried flowers) · 70 x 20 x 25 cm (WxHxD) · 2016
A Sense of Touch
Exhibited at Galletly Gallery, New Hampton · Photography · 38 x 31 cm (WxH) · 2016
The City Beautiful
Exhibited at Kharkiv Local Gallery, Ukraine · Mixed (styrofoam, paper, white threads) · 30 x 50 x 25 cm (WxHxD) · 2015
Spatial Rhythm
Exhibited at Kharkiv Local Gallery, Ukraine · Mixed (styrofoam, paper, white threads) · 66 x 50 x 36 cm (WxHxD) · 2015
What if?
Exhibited at Kharkiv Local Gallery, Ukraine · Mixed (styrofoam, paper, white threads) · 58 x 50 x 36 cm (WxHxD) · 2015