News reporting focused on AI and machine learning, covering the companies behind these technologies, their real-world applications, and the ethical concerns they raise. This includes areas like generative AI (large language models, text-to-image and video), speech tech, and predictive analytics.
Xiaomi released Xiaomi-Robotics-1, an AI model for robots that improves with more training data. To build a large dataset, the team used handheld grippers instead of physical robots, collecting over 100,000 hours of motion recordings. Tests showed that adding data boosted performance far more than increasing model size, with success rates in unfamiliar environments rising from about 25 percent to 75 percent.
Kaikaku.AI introduces Epicure, three nearly identical AI models trained on different data sources. One learns from recipes, another from flavor molecules, and a third blends both. The models reveal how training data choice changes answers to simple questions like what goes with chicken.
Researchers Nick Levine, David Duvenaud, and Alec Radford released talkie, a 13B language model trained on historical English text before 1931. The base version used 260B tokens, while the instruction-tuned model powers a chat interface. Both models carry an Apache 2.0 license, with training data free from copyright restrictions.
Researchers measured a 'flinch' in seven pretraining datasets from five labs, where models assign far lower probabilities to charged words like 'deportation' compared to neutral ones. Even models labeled uncensored, such as heretic based on Qwen3.5-9B, show this effect and worsen slightly after refusal ablation. Profiles vary by lab and year across categories like slurs, violence, and political terms.