Can Australia move up the AI value chain? These founders think so.
On Wild Hearts with Pip Bingemann and Tom Humphrey: building specialised models, open-source research and the importance of fast inference.
Podcasts, articles, tutorials, experiments and press coverage.
On Wild Hearts with Pip Bingemann and Tom Humphrey: building specialised models, open-source research and the importance of fast inference.
On Wild Hearts: lessons from Amazon and Stitch Fix, what “developer first” means, and the decision to make Marqo open source.
A conversation on the path from physics to building Marqo, and on multimodal vector search.
Introducing Marqo-FashionCLIP and Marqo-FashionSigLIP: 22% and 57% higher average text-to-image Recall@1 than FashionCLIP2.0, respectively, across six fashion datasets in the release benchmark.
Inferring warehouse layouts using ideas from recommender systems, distance geometry and structural biology.
With Dave Spiegel: applying optimisation to office seating while accounting for people’s needs and preferences.
Exploring reinforcement learning as a way to solve travelling-salesman routing problems.
Using text and images together for search: multi-part and multimodal queries, searching by prompting, and promoting or suppressing content by theme.
Using Marqo as editable external memory for product question answering and conversational agents.
Adding localisation and open-vocabulary reranking to image retrieval, with working code.
Exploring generated images through semantic search, filtering and labelling with Marqo.
Reporting by other publications.
Pure AI on the launch of Celeris-1, which uses diffusion-based generation instead of autoregressive decoding for faster responses.
Amazon Science on reusable perception models that help new robotic systems learn package-handling tasks with less task-specific data.
Amazon Science covers the framework for retraining robotic systems as their operating environment changes.
Amazon Science examines the perception and learning systems behind Robin’s package handling.