Batch size is the number of training examples a model processes before updating its parameters. A small batch uses less memory and produces noisier gradient estimates, which can sometimes help a model generalize, but it may take longer to use hardware efficiently. A large batch offers steadier updates and better parallel processing while requiring more memory and often careful learning-rate adjustment. The best value depends on the model, dataset, optimizer, hardware, and training objective. Batch size also affects how frequently parameters change during an epoch, so comparisons should account for the total number of examples processed rather than treating the setting as an isolated performance switch.
Pinterest is turning to open source AI models to cut expenses and enhance its visual search and shopping tools, even as the company forecasts softer holiday revenue. CEO Bill Ready says the approach will support performance while keeping infrastructure costs in check.
Elon Musk has unveiled Grokipedia – an AI-driven encyclopedia built on Grok that aims to offer factual, bias-free knowledge as an alternative to Wikipedia.
Andrew Tulloch, co-founder of the $12 billion AI startup Thinking Machines Lab, has joined Meta after previously rejecting what reports described as a $1.5 billion offer — a figure Meta has since called ‘inaccurate and ridiculous.’
OpenAI has announced a strategic partnership with Broadcom to design and deploy 10 gigawatts of custom AI accelerators, expanding its hardware ecosystem beyond Nvidia and AMD.
Google has unveiled Gemini Enterprise, a next-generation AI platform designed for business productivity and collaboration, positioning itself directly against Microsoft’s Copilot and OpenAI’s enterprise solutions in the race for workplace AI dominance.
Google Cloud and Figma have launched an expanded partnership to embed Google’s generative AI models – including Gemini 2.5 Flash and Imagen 4 – directly into Figma’s design tools, reducing latency and accelerating creative workflows.