10 results on this page
An independent builder pretrained a 1.02-billion-parameter Kimi-K3 replica on 5 billion decontaminated tokens for roughly $250. The model achieved 33.4% on HellaSwag, exceeding the reported 28% score of GPT-2 124M despite receiving no instruction tuning.
New releases from Chinese competitors are narrowing the gap with Silicon Valley’s AI products, including ChatGPT and Gemini. Lower pricing is helping these services gain market share, alongside offerings such as DeepSeek and Kimi agents.
Kimi K3 and Qwen3.8-Max have reportedly introduced revenue-sharing clauses into their open-source strategies. The shift suggests that open models may be moving from unrestricted free use toward a freemium model.
Serval has made Catalyst generally available as an administrative super agent that discovers repetitive work and builds governed enterprise automations. It can also create background agents that monitor connected systems, identify emerging issues, and suggest fixes before employees submit tickets.
MiniMax Design uses an Agent to orchestrate multiple models for professional content production. Its positioning reflects a broader shift in AI video competition from model capabilities alone toward integrated products and measurable commercial value.
Researchers propose FAR, a human-AI discovery pipeline that finds open problems from research literature, attempts solutions, and recommends promising results for expert review. In a combinatorics pilot, it narrowed 5,245 papers to 77 items for author-team evaluation and surfaced multiple potential mathematical discoveries.
The US Trade Representative’s reference to “digital trade alignment” in talks with Prime Minister Mark Carney’s government is raising concerns in Canada. Critics worry the agreement could limit Canada’s ability to independently regulate technology companies and AI.
MiniMax launched MiniMax Design, a production workflow built around its H3 video model. The platform organizes professional video capabilities into executable nodes for continuous editing, collaboration, and delivery across video, image, music, and voice generation.
Apple Machine Learning studies how to mix scarce target-domain data with abundant generic data during language-model pretraining. Across more than 2,000 training runs, the research examines the trade-off between insufficient target exposure, repeated examples, diminishing returns, and overfitting.
Gemini Live can now start Deep Research reports through voice commands. It can work on reports in the background and let users discuss the findings when they are ready.