ToolGrad: Efficient tool-use dataset generation with textual “gradients”
Machine Intelligence
-
-
Meet Redis LangCache: A Managed Semantic Cache That Cuts LLM API Costs by Up to 90% and Returns Cache Hits Up to 15x Faster
Production LLM purposes not often obtain a query no person has requested earlier than. Support assistants and RAG pipelines area the identical intents 1000’s of occasions a day, every phrased in another way, and most stacks deal with each phrasing as a contemporary, totally billed request. Redis LangCache is a completely managed semantic caching service…
-
NVIDIA Details BioNeMo Inference Runtime (BioIR): 2.90x Higher Boltz-2 Folding Throughput and 58.5K Residues per GPU-Hour on 8xH100
Biomolecular construction prediction has shifted from single-target runs to proteome-scale worklists. The bottleneck is not whether or not a mannequin can fold a protein. It is how briskly a complete queue of unbiased targets strikes by parsing, featurization, GPU inference, and output writing. NVIDIA’s new technical deep dive walks by BioNeMo Inference Runtime (BioIR), a…
-
OpenAI Launches the Agents API in Public Beta, Putting the Codex Harness Behind One API Call
OpenAI has launched the Agents API in public beta. It offers builders the identical harness and infrastructure that run Codex. OpenAI hosts and maintains the harness. Developers run the agent’s compute in an OpenAI-managed sandbox, their very own infrastructure, or a associate sandbox. Is it deployable? Yes. It is stay for all builders in public…
-
Supply chains detect fast, act slow: How AI agents fix it
Supply chain disruption value companies about $184 billion in 2025, in keeping with the J.S. Held Global Risk Report, and most of that invoice nonetheless buys quicker detection, not quicker motion. That determine is often handled as climate (i.e. storms occur, prices observe.) Treated as a product specification as an alternative, it highlights an working…
-
7 signs your AI infrastructure is still stuck in the HPC era
A GPU cluster can look completely wholesome on a monitoring dashboard whereas producing little or no of worth, and most enterprises are looking at precisely that chart in the present day. Roughly two-thirds of organizations report peak GPU utilization beneath 70%, in response to a State of AI Infrastructure at Scale report cited in a…
-
Building a Reliable Foundation for Agentic AI in SMBs
This article is sponsored by Salesforce and was written, edited, and revealed in alignment with our Emerj sponsored content guidelines. Learn extra about our thought management and content material creation providers on our Emerj Media Services page. Customer-facing organizations now face a widening capability hole pushed by escalating multi‑channel demand and the constraints of human-only…
-
JD.com expands physical AI in logistics with 3 million robots
JD.com is increasing AI and robotics throughout its logistics community beneath a brand new Physical AI Acceleration Plan, whereas reiterating a five-year goal to acquire 3 million robots, 1 million autonomous autos, and 100,000 supply drones. The firm launched the plan at JDDiscovery 2026 in Beijing. JD Logistics additionally unveiled its industrial Wolf Robot sequence,…
-
Running AI in Production: Reliable AI Workflows, Governed Access, at Scale
Most groups constructing AI workflows attain for brokers and abilities, and find yourself with one thing probabilistic from finish to finish. The agent calls no matter device it decides to name, reaches for an costly mannequin on a step that wanted a lookup, and runs as a black field: exhausting to validate, exhausting to foretell,…
-
DeepSeek AI Released DeepSeek-V4.1-Flash with 1M Context, FP4 KV Cache, and Cross-Layer Attention Reuse
Long-horizon brokers have turned LLM serving into an input-heavy workload. Repeated prefills and million-token contexts depart KV caches that pressure HBM, SSD capability, and bandwidth. DeepSeek AI constructed its latest launch round that actual bottleneck. DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts mannequin with 552B spine parameters, 196B further Engram parameters, and a 1M-token context window. It…
