Business
OVHcloud Frontier AI Models: The €200M Bet to Rival Mistral
OVHcloud frontier AI models: why Europe's biggest cloud firm wants to be the No. 2 LLM maker behind Mistral, what…
Topic
Large Language Models
Science
Autonomous Construction: A Crane That Builds and Reports Back
Autonomous construction gets a voice: UBC researchers pair reinforcement learning with ChatGPT 5 so a robot crane builds walls and…
Science
Synthetic Polling: AI Personas Missed a 13-Point Approval Slide
Synthetic polling held flat at 46% approval while real voters fell to 34%. Pew's AI persona test shows why chatbots…
Science
Synthetic Survey Respondents: How PARADE Cuts Error 52.8%
Synthetic survey respondents from 107 plug-in LoRA blocks: how Tsinghua's PARADE cut distribution error 52.8% and where the method still…
Newsbeat
MiniMax M3.1 Flash: The $0.15 Agent Model That Was Space Bunny
MiniMax M3.1 Flash costs $0.15 per million input tokens and was secretly OpenRouter's Space Bunny. Benchmarks, pricing and the catches,…
Science
Relearning Savings: Why AI ‘Forgotten’ Facts Return 18% Faster
Relearning savings study: a Qwen2.5-1.5B model 'forgot' facts, then relearned them about 18% faster. Here's why accuracy scores miss what…
Science
LLM Calibration Agent: 9 Simulator Runs vs Nelder-Mead’s 17
An LLM calibration agent cut costly simulator runs from 17 to 9 in a new Rutgers study. See where physics-labeled…