llama.cpp distributes inference across heterogeneous devices via ggml RPC backend

A French student built Pamk, a 3D-printed desk robot on an ESP32-S3 that manages tasks and reminders via the custom wake word "Hey Pamk," trained on thousands of the builder's own recordings. It needs no companion app.

AT&T's inhouse legal agency LegalEdge, launched in January, has handled nearly 100 matters in its first six months. OpenAI's Jason Boehmig said the effort centers lawyers' judgment, following OpenAI's Astra for Law launch and legal-tech plugin collection.

Latent Space podcast interviews Periodic Labs' Liam Fedus and Ekin Dogus Cubuk on "synthesis superintelligence" — reinforcement learning grounded in physical experiments and autonomous labs. Periodic launched last September and is now building AI systems that reason over noisy physical experiments.

The New Stack webinar on October 22 features Traversal co-founder and CTO Raaz Dwivedi on why most AI SRE tools stop short of deciding when an incident needs attention. Dwivedi is also an assistant professor at Cornell Tech.

AWS published guidance on partitioning expensive GPU clusters among multiple internal teams using Amazon SageMaker HyperPod, targeting generative AI workloads such as LLM training. The approach covers isolation boundaries, resource fairness, and operational independence for teams sharing the same hardware.

Mellum2.1 is a 12B mixture-of-experts thinking model activating 2.5B parameters per token, shipping under Apache 2.0 on Hugging Face with a 131,072-token context. It beats Mellum2 on 15 of 17 benchmarks and scores 82.0 on LiveCodeBench v6, but trails Qwen3.5-9B on Terminal-Bench 2.1 (17.4 vs 21.7).