Infrastructure Correspondent
Adrian Cole follows the infrastructure behind the industry: chips, data centers, power and the economics of compute. His coverage asks what systems cost to build and operate, where the bottlenecks sit and which performance claims survive a like-for-like comparison.
905 published articles · Showing 21–40
AllenAI's StateBench establishes a four-tier hierarchy for state tracking and shows that even frontier models fail on the hardest variants as inputs lengthen....
Read article →The plan puts a China-first hardware bundle on the international trade-show circuit. ISHO's release names CES 2027 as the vehicle for its expansion into international markets, with no venue, booth or pricing details given....
Read article →The papers touch the resources that set training and inference economics — compute time, memory, communication and measurement shots — rather than introducing new model capabilities....
Read article →OpenAI released GPT-6.1 Terra with API pricing at $1 per million input tokens—half the cost of its predecessor—targeting cost-sensitive enterprise automation and computer-use tasks....
Read article →A preprint proposes a physics-aware self-supervised learning method that learns from unlabeled Raman spectroscopy data, potentially reducing reliance on costly labeled datasets and improving model transferability....
Read article →The framework uses a dual‑memory architecture to inject expert knowledge and recent task history into an iterative LLM‑driven optimization loop, aiming to make GPU kernel tuning more autonomous....
Read article →The paper's main methodological claim is that team scores conflate generation and selection, and delivery rules can reverse apparent topology advantages....
Read article →SAGE-CP, a calibration method in an arXiv preprint, guarantees coverage within each computational route of adaptive VLM ensembles, raising worst-route coverage from 0.623 to 0.887 while retaining efficiency gains....
Read article →AWS Trainium gains a native PyTorch backend, TorchNeuron, eliminating the need for custom Hugging Face integration code and matching the developer experience of GPUs....
Read article →Researchers propose adapting labeled random finite sets, a concept from multi-object tracking, to manage the inherent uncertainty in geotechnical databases, potentially strengthening data-driven infrastructure analysis....
Read article →A theoretical preprint extends the formal understanding of in-context learning to unsupported instructions and tasks that can be split into pretraining-supported subtasks, though no experiments validate the claims....
Read article →The technique relies on concurrent rewards and a shared token budget to cut compute cost without explicit length penalties....
Read article →A preprint proposes a causal‑beam method to transfer task knowledge from human videos to robot manipulation without direct motion mapping, reporting improved success over baselines. The work has not been peer‑reviewed or independently tested....
Read article →A new arXiv preprint introduces DialFork, a benchmark that isolates procedural hallucination—conversations that remain fluent but drift from their intended process—across 36,278 dialogues in 12 domains....
Read article →Google DeepMind is hiring a nuclear physicist or engineer to assess how its AI models could be misused for weapons proliferation, part of its Responsible Development and Innovation team....
Read article →The ruling gives Apple expedited discovery in a trade-secret lawsuit that could influence a hardware rivalry as OpenAI develops a device that may compete with the iPhone. Apple sued OpenAI in July 2026, alleging a pattern of trade-secret theft....
Read article →The push comes as cloud providers seek returns on massive AI infrastructure investments. Anthropic launched the Claude Frontier Academy, a training program backed by $100 million in initial funding, according to a company announcement carried by Channel Dive....
Read article →The move extends a March partnership and targets the growing multimodal demands of creative professionals, with NVIDIA’s RTX Spark chip central to the speed claims....
Read article →The suits, filed by a bipartisan group of attorneys general, accuse OpenAI of designing ChatGPT to maximize engagement among minors while concealing risks of harmful outputs. OpenAI says it has built age-specific safeguards and will work with states on safety policy....
Read article →French entities are deploying agentic computing on Nvidia hardware from orbit to public-sector offices, turning two years of summit pledges into sovereign infrastructure that includes local server manufacturing and a $1 billion satellite constellation....
Read article →