AI companies use model specifications to define target behaviors during training and evaluation. Does the current specification describe the expected behavior accurately enough, and do cutting-edge models exhibit different behavioral characteristics under the same...
In this tutorial, we build from scratch a high-level computer usage agent that can reason, plan, and execute virtual actions using a local open weight model. We created a miniature simulated desktop, equipped it...
Liquid AI has released LFM2-VL-3B, a 3B parameter visual language model for image-to-text tasks. It extends the LFM2-VL family beyond the 450M and 1.6B variants. The goal of this model is to improve accuracy...
In this tutorial we will explore Special servicea lightweight and powerful serving framework that allows us to deploy machine learning models as APIs with minimal effort. We build and test multiple endpoints to demonstrate...
A team of Salesforce AI researchers has launched WALT (Web Agent for Learning Tools), a framework for reverse engineering potential website functionality into reusable callable tools. It re-architects browser automation around callable tools instead...
Open vocabulary object detector answers text queries with boxes. In remote sensing, zero-shot performance degrades because the class granularity is very fine and the visual environment is unusual. The Google research team proposed flamea...
Computer usage of agents is limited to primitives. They click, type, scroll. Long action chains amplify grounding errors and wasted steps. Apple researchers launch UltraCUAa fundamental model for building hybrid operation spaces that enable...
In this tutorial, we’ll explore how to build a small but functional multi-agent system using: uAgents frame. We set up three agents—catalog, seller, and buyer—that communicate via a well-defined messaging protocol to simulate real-world...
Anthrogen launches Odysseya family of protein language models for sequence and structure generation, protein editing, and conditional design. Production model parameters range from 1.2B to 102B. Anthrogen’s research team positions Odyssey as a cutting-edge...
Puji Artificial Intelligence Already open source PokeeResearch-7Ba 7B parameter deep research agent that performs a complete research loop, decomposes the query, issues search and read calls, validates candidate answers, and then synthesizes multiple research...