Research
Foundation & multimodal models
Learning objectives, adaptation, language, vision and the relationships between modalities.
- How are model improvements evaluated?
- Which assumptions and limitations matter?
AI Progress Summit
Topics
A connected technical scope, from how models learn to how systems behave in use.
Evaluation
Evaluation & reproducibility
Benchmarks, human evaluation, robust experimental design and reproducible evidence.
- What does a metric actually measure?
- Can a result be reproduced in a different setting?
Research
Agents & tool use
Planning, tool interaction, retrieval and the evaluation of systems that act across multiple steps.
- How is a complete task evaluated?
- Where do failures occur across a workflow?
Deployment
Deployment & AI systems
Model serving, monitoring, lifecycle management and integration into real workflows.
- How does behaviour change after deployment?
- What evidence supports a release decision?
Deployment
Efficient inference & compute
Latency, throughput, memory, energy and the tradeoffs involved in operating AI systems.
- Which constraints matter for the application?
- How are performance tradeoffs measured?
Research
Data & learning methods
Data quality, curation, learning efficiency and methods that connect training choices to outcomes.
- How does data affect a result?
- Which comparisons make a method credible?
Evaluation
Responsible & reliable AI
Robustness, uncertainty, failure analysis and the responsibilities that follow from system use.
- What happens outside the expected conditions?
- How are limitations communicated?
Deployment
Applied AI & real-world progress
Application context, user needs and useful outcomes across scientific, industrial and everyday settings.
- What changed for the people using the system?
- How is useful progress distinguished from a model score?
Call for papers
Connect a research question with a method, evidence and honest limitations.