New memory architecture could enable multi-TB GPU capacity

Basel Action Network's new estimate is far higher than prior studies because it counts all data center support infrastructure, not just servers. Less than a quarter of the 68.3 million tons of e-waste generated worldwide each year is formally collected and recycled.

NVIDIA's TensorRT Edge-LLM ran Qwen3.6-27B on a single Jetson AGX Thor Developer Kit at 52.33 tokens/sec, finishing all 1,007 turns in 24 minutes 36 seconds versus llama.cpp's 2 hours 37 minutes. It uses NVFP4 quantization, tree-based multi-token prediction, and KV cache reuse.

LARA (Lightweight Additive Residual Adaptation) is a research project and PyTorch library for making post-training modular on frozen language models, letting small composable behaviours be added without touching base weights.

OLogic CEO Ted Larson will lead "Egocentric Robot Learning: Teaching Robots Through Human Experience" at RoboBusiness, October 20-21, 2026, in Santa Clara, Calif. The session starts at 10:45 a.m. on day two and covers first-person human demonstrations via wearable cameras and teleoperation.

Ads on Reddit recruiting Latin American voices for commercial AI voice-cloning services ask applicants for a voice sample plus name, phone number and email. The poster argues that combination enables impersonation and scam risk.

Glyph pairs a Descriptor agent that grounds column descriptions in pipeline source code pulled from enterprise GitHub with a Tagger that fuses three strategies over a 275-leaf Data Classification Ontology. A fine-tuned 6-layer MiniLM encoder lifted same-tag retrieval from NDCG@10 0.55 to 0.92.

OpenAI's ChatGPT Admin Console now surfaces AI adoption, usage, and spend across an organization, plus Codex contribution data. A ChatGPT Admin plugin turns those analytics into actions.