Latest in AI
Every brief, newest first. Each one links to its primary source.
-
AWS ports DeepSeek’s DeepEP to its own fabric for MoE training
Amazon’s upstream work moves DeepEP off CUDA-specific RDMA, and it reports 40 percent more rollout throughput across 48 instances.
-
Cohere puts Compass retrieval in the cloud, in private beta
Compass becomes a managed service for a handful of beta partners, with no pricing and a benchmark win measured on Cohere’s own dataset.
-
Nvidia starts qualifying batteries and coolers for AI sites
Nvidia’s new DSX Ready programme certifies batteries and cooling units for AI sites. One category is reviewed by Nvidia, the other self-attested.
-
EU adopts a sustainability rating for data centres above 500kW
Brussels has adopted a rating scheme for data centres above 500kW, with first labels due in 2027 and a consultation on minimum standards now…
-
Microsoft opens up a retrosynthesis model chemists preferred
Microsoft has published its retrosynthesis model in Nature and released the weights under MIT. Blind-test chemists preferred its route suggestions.
-
Grok 4.7 competes on price, not on top scores
SpaceXAI holds Grok 4.6 prices for its new flagship, which trails Fable 5.1 on most of the benchmarks it published. The pitch is cost…
-
OpenAI claims 100 solved maths problems, and appoints watchers
OpenAI says an internal model has resolved over 100 open maths problems, and names nine mathematicians to advise on how such results are released.
-
Google’s Pixel Watch can now call an ambulance by itself in Europe
Breathing Emergency Detection escalates on its own, and Google publishes no accuracy figures and names no countries.
-
Perplexity’s local agent runs on AMD Ryzen AI Max PCs
Running inference locally lets agents skip metered cloud credits, but AMD names no models, no figures and no supported systems.
-
Perplexity replaces its search backend with an in-house engine
Photon cut estimated cost per task by 68%,, Perplexity says, and it labels its own latency figures self-reported.