Large language models (LLMs) have become a common tool in most investigative newsrooms. But what do you do when prudish language models refuse to process what you are investigating? Or you have so much sensitive information that it makes your stomach hurt to send it to Big Tech?
Enter local AI models!
Norwegian Broadcasting Corporation (NRK) has used AI models running on their own machines or leased ones to carry out several projects. Together with Lighthouse Reports, we exposed a hidden class divide in Norwegian courtrooms – revealing that the wealthy receive more lenient sentences than the poor – by analyzing 9,000 verdicts. The NRK team then turned its attention to the adult industry, reviewing over 1,000 films to document a sharp rise in choking incidents.
While the subjects differ vastly, the projects share a common thread: the controlled use of local AIs to process massive datasets. We want to share our methodology, our findings, and insights along the way – but most of all demonstrate how to get you started with using the latest local models for your specific investigative needs.
In the presentation, we will show concrete examples of how to run the latest local models and invite the audience to think with us about how these models can enhance investigative workflows.
