Results 721 - 730 of 24387

Machine learning (ML) and artificial intelligence (AI) are not standalone artifacts: they are ecosystems where foundation models are adapted and deployed through layered pipelines spanning developers, platforms, users and regulators. This talk explores how the structure of these ecosystems shapes the distribution of value and risk, and system-level properties like safety and fairness. I begin with a game-theoretic model of the collaboration between general-purpose producers and domain specialists, using it to examine how participation incentives and regulatory design shape equilibrium behaviors. I then connect these formal insights to empirical measurements from 1.86 million open-source AI models, reconstructing lineage networks to quantify how behaviors and failures propagate through fine-tuning processes. I close by outlining my research agenda aimed at building the technical and policy foundations needed to steer collaborative AI ecosystems toward robust, accountable and democratic outcomes.

Nearly all scaling laws research treats English as the evaluation objective. In this work, we redefine scaling laws to support multilingual objectives, and to inform how practitioners should mix hundreds of language sources to optimize for a non-English language target. Our state-of-the-art scaling laws are also used to determine if its better to pretrain from scratch or finetune from a multilingual checkpoint, and how many languages different model sizes can easily support. We release our cross-lingual transfer matrix for practitioner use and downstream analysis.

Mechanisms for continued self-improvement of language models without external supervision remain an open challenge. In this talk, I will present Peer-Predictive Self-Training (PST), a label-free fine-tuning framework in which multiple language models collaboratively improve by using a cross-model-aggregated response as an internal training target. Given a prompt, models generate responses sequentially; a final aggregated answer—often more reliable than any individual response—then serves as the target for learning. To guide updates, PST measures how informative each intermediate response is about the aggregate using pointwise mutual information (PMI) and scales gradient updates accordingly: responses already aligned with the aggregate are updated less, while misaligned or less informative responses are updated more. I will conclude with empirical results on mathematical reasoning tasks and discuss when peer-based self-training helps.