Events

For decades, the internet’s business model has been built around search: websites get indexed by search engines and monetize the human traffic that search sends to them. With the rise of large-language-model agents, this model is breaking. AI systems increasingly fetch web content directly in order to answer questions or perform tasks for users. When AI agents replace human visits, websites lose the traffic that historically generated revenue.

This shift creates the need for new business models for online content—models in which AI agents pay websites for the content they consume, whether for grounding, retrieval, or training. One emerging approach is Pay-Per-Crawl (PPC), where websites charge AI models for access to their pages. Companies like Cloudflare have begun building the infrastructure for this market, but the economics remain entirely undeveloped.

Our project designs the market mechanisms that should govern PPC. We evaluate pricing rules—fixed-fee crawls, content-sensitive pricing (e.g., politics vs. sports), crawler-identity pricing, and hybrid mechanisms—using three criteria: (1) revenue for content creators, (2) minimal degradation to AI answer quality relative to a zero-price world, and (3) design simplicity.

Requirements: strong interest in market design for emerging AI economies; comfort with ML methods such as embeddings-based content valuation; ability to run simulations of multi-sided markets; and willingness to engage with cutting-edge industry questions involving Cloudflare, OpenAI, and others.

Presented by Joel Rozowsky

The Gerstein lab conducts computational biology & bioinformatics research in the biomedical and genomic fields. We use various computational analytics methods including artificial-intelligence / machine-learning techniques to analyze large biomedical datasets and develop bioinformatics tools. The lab has particular focuses on the following areas of research: neurogenomics, personal genomes, genomic privacy and genome annotation.

• AI / ML: Multi-modal representation learning, graph neural networks. (No prior biology background required).
• Coding: Python.
• The Challenge: Learning robust representations for high-dimensional, noisy, and heterogeneous spatial omics data.

In the United States, police regularly use force against civilians. Much of the prior research in this area has focused on lethal force, but much less is known about the vast majority of incidents that do not result in death. This project aims to generate the first national estimate of the annual number of use-of-force incidents in the US overall and by race/ethnicity. Leveraging newly available use-of-force data for thousands of agencies and a national dataset of predictors, we use Bayesian and machine learning approaches estimating use of force counts for the entire US.

We have data covering large areas of the sky, which needs to be scanned automatically to find candidates for merging galaxies, dual AGN (Active Galactic Nuclei, which is when both black holes in the merging galaxies are growing rapidly and thus shining brightly), and all combinations in between. Right now, we’ve trained a first-generation CNN that does the job okay, but at a minimum, we need to update it to incorporate multiple images taken at different wavelengths, to query the full data set (right now, we start with known AGN and look for a companion), and to distinguish stars from AGN. More ambitious goals: Is there a better approach than training a CNN (we use a large number of simulated images and then a more limited number of bona fide cases)? Can we train it more efficiently and/or improve the code in other ways?