Curiosity,
built.
A design & development studio turning good questions into apps, tools, and things that work.
Explore the research
01 / Algorithmic Research Group
Research,
in the open.
We study how AI systems behave, what they learn, and where they fail. Experiments, benchmarks, and open tools from the research practice at ARG Studio.
Featured research / Forthcoming
Causal Tests of Language-Model Reports
Changing a learned preference moves the model’s self-report. But the report captures only about 36% of the behavioral change.
Read the research story
The research collection
Every study. Open to explore.S2ORC CS Enriched
A filtered and LLM-enriched version of Allen AI's S2ORC corpus containing 1.1 million computer science papers with structured extraction of methods, models, datasets, metrics, compute usage, and limitations added to every row.
Read the researchGPU Kernel Optimization
A retrospective on 131,520 GPU kernel optimization attempts that were invalidated when agents were found to be substituting high-level PyTorch API calls instead of writing actual kernels.
Read the researchNeural Architecture Search
A tiny recursive reasoning model trained to rank architectures by predicted performance achieves 8-10x sample efficiency over random search and transfers zero-shot across datasets with minimal loss in ranking quality.
Read the researchARIA Benchmark
A suite of five closed-book benchmarks probing the ML knowledge that frontier language models have internalized during training.
Read the researchArXiv Research Code
A collection of 4.7 million code files from 129K research repositories linked to arXiv computer science papers.
Read the researchArXivDLInstruct
A dataset of 778,152 functions extracted from arXiv-linked research code, each paired with instruction prompts, for training ML-specialized code generation models.
Read the researchDeltaML-Bench
A 48-task benchmark for evaluating autonomous ML experimentation in real research repositories.
Read the researchTeaching Models to Bluff
We implemented five LLM agents playing the social-deduction game Secret Hitler with structured logging to quantify deception, belief accuracy, and coalition dynamics.
Read the researchML Research Benchmark
A benchmark suite of 7 competition-level ML challenges for evaluating whether AI agents can perform genuine research iteration beyond baseline reproduction.
Read the researchCode, datasets & models
13 open resourcesCode & Benchmarks
Datasets
A few things
we’ve put into the world.
Private, local software. Publishing without a gatekeeper. Creative work that follows its own rules.
saylightSaylight
Private dictation with local speech recognition. No uploaded recordings and no cloud transcription.
Harness
Local desktop controls that make AI-agent work visible, bounded, and interruptible.
presskeepPresskeep
Signed, portable publishing that does not depend on one platform’s permission.
ELSEWHERE
Search apps, files, notes, and clipboard history without uploading queries, filenames, or an index.
Dropbeam
Send files with a one-time code and an encrypted direct-or-relayed connection—without accounts or cloud folders.
Experiments & collaborations
The stranger side.
Motion becomes music. Data becomes sculpture. Made with artists and researchers, including Alex Murray-Leslie of Chicks on Speed.
2015
Footwerk
With Alex Murray-Leslie of Chicks on Speed, recorded movement becomes visual trails and synthesized sound. Exhibited at the Venice Biennale and ArtScience Museum.
With Alex Murray-Leslie (Chicks on Speed), Penn State Biomechanics Lab
Shown at 56th Venice Biennale · ArtScience Museum, Singapore
2018
THE_OPER&
Machine learning becomes a performer in an algorithmic opera at Duke. My contribution made image classification probabilities visible as a shifting field of interpretations.
With Bill Seaman, John Supko, Jim Findlay, Lorelei Ensemble, Keith Scretch
2014
Biobehavioral Landscapes
Emotional life, given terrain. Data from 150 adults becomes a series of statistical landscapes, modeled in Grasshopper and milled from oak.
With Nilam Ram, Penn State Department of Human Development
03 / The practice
Part design studio.
Part engineering lab.
Always making.
ARG Studio builds software that keeps data and control close to the people using it. Other questions become sculptures, performances, or research. The medium changes; the independence does not.
Meet the studioDesign
Clear interfaces. Considered interactions.
Develop
Apps, tools, and the systems behind them.
Explore
AI experiments. Questions worth testing.
Follow the
interesting questions.
Strange instruments. Software with a point of view. Working ideas and the stories behind the things we make.
Design the Handoff
Human review is part of the product. Design the evidence, decisions, and recovery path with the same care as the automation.
Read the story
Sculpting with a flock
Particles leave geometry in their wake. An experimental CAD tool turns movement into forms you can fabricate.
Inside the projectThe Two-Minute Task
Small tasks accumulate into real work. Follow the repetitions, interruptions, and queues to understand what they cost.
Read the story
Listening to 400,000 years of ice
With Mark Ballora and the Penn State Polar Center, Antarctic data becomes a composition for six synthesized voices.
Inside the projectA hundred days online, scored for radio
Five instruments turn a radio host’s web activity into a musical portrait for To the Best of Our Knowledge.
Inside the projectThe Work Before the Build
A project gets easier to build when its owner, interfaces, test cases, and operating assumptions are already on the table.
Read the story