Latest research
October 2, 2024
Investigating pretraining dynamics and stability with OLMo checkpoints
We use data from our open pretraining runs to test hypotheses about training dynamics in OLMo checkpoints.September 4, 2024
OLMoE: An open, small, and state-of-the-art mixture-of-experts model
Introducing OLMoE, the first model to be on the Pareto frontier of performance and size, released with open data.August 12, 2024
Digital Socrates: Evaluating LLMs through explanation critiques
Digital Socrates is an evaluation tool that can characterize LLMs' explanation capabilities.August 8, 2024
Open research is the key to unlocking safer AI
Ai2 presents our stance on openness and safety in AI.July 3, 2024
Broadening the scope of noncompliance: When and how AI models should not comply with user requests
We outline the taxonomy of model noncompliance and then delve deeper into implementing model noncompliance.June 24, 2024
PolygloToxicityPrompts: Multilingual evaluation of neural toxic degeneration in large language models
New research on AI prompt toxicity, revealing insights into neural toxic degeneration across diverse languages.May 16, 2024
Data-driven discovery with large generative models
We believe AI can assist researchers in finding relevant preexisting work to expedite discoveries. Here's how.April 23, 2024