Introducing a new dataset in the SWE-bench family with 300 curated tasks in 9 programming languages to evaluate LLMs on software engineering tasks.
Creating a small benchmark to test how well multimodal language models can find specific objects in complex Where's Waldo-style illustrations.
Exploring a missing piece of the software observability stack to monitor business logic.
Implementing partial builds to reduce the time taken to live reload a static site by >99%.
Understanding how the UK raises and spends money, drawing from Paul Johnson's book Follow the Money.