← Back to events

Stay in the know

Connect with the Snorkel team at the booth or at our side events.

Expo talk

The Art & Science ofBenchmarking Agents

Dec 6, 12:00–1:00 PMGrand Ballroom B1

Benchmarks matter more than ever as agents take on real work, yet benchmaxxing, benchslop, and reward hacking are eroding trust in them. This talk argues that a great agent benchmark is both a research agenda (the art) and a trustworthy measurement instrument (the science), using Senior SWE-Bench as a case study. Attendees leave with a practical checklist for building agent benchmarks that stay difficult, fair, and useful.

Kiran Lakkaraju

Kiran Lakkaraju

Director of Research, Frontier Benchmarks

Diagram with three axes: Autonomy Horizon (how independently can the agent operate), Environment Complexity (how dynamic is the operating environment), and Output Complexity (how sophisticated is the deliverable)

Accepted papers

Five papers fromour research team

SkillOrchestra

Routing tasks by agent competence and cost

Poster presentation · Wed, Dec 9 · 10:00 AM · Session 3

Snorkel authors

Frederic Sala

arXiv

Continual Learning Bench

Do agents improve with experience?

Poster presentation · Thu, Dec 10 · 10:00 AM · Session 5

arXiv

Agents’ Last Exam

Long-horizon professional tasks, <1% average pass rate on the hardest tier

arXiv

JudgmentBench

Expert attorney preference judgments for ranking legal AI output

Snorkel authors

Charles Dickens

arXiv

SlopCodeBench

Agent-written code gets more bloated and eroded with every iteration

Snorkel authors

Frederic Sala

arXiv

Side events

Find us beyondthe expo hall

Happy hour

NeurIPS Skyline Happy Hour

Drinks and harbour views. Limited space, so request an invite.

Wed, Dec 9 · 6:00–9:00 PMThe Oriana Sky Tower, Circular Quay
Request invite

Workshop

Agentic AI Benchmarks and Applications for Enterprise Tasks

The 2nd workshop on benchmarking agentic AI for real enterprise workflows.

Fri, Dec 11Sydney, Australia
Workshop site

Workshop

Who Verifies the Agents?

Making verification a first-class problem for building reliable AI agents. Co-organized by Snorkel’s Ramya Ramakrishnan.

Sat, Dec 12Sydney, Australia
Workshop site

The team

The people behind the research

Vincent Chen

Founding Research Fellow

Frederic Sala

Chief Scientist

Kiran Lakkaraju

Research

Jake Rudulph

Research

Charles Dickens

Research

Rohit Deshpande

FDR

Nathan Xu

FDR

Zach Kleinbaum

FDR

Keith Cannon

FDR

Connect with us

Stay in the know

Request an invite to our Skyline Happy Hour, or get notified about other side events.