Google launched Gemini 3 Flash as its new default AI model, offering up to three times faster performance than Gemini 2.5 Flash while being more cost-effective. PCWorld reports the model excels in multimodal tasks, scoring 81.2% in MMMU-Pro benchmarks and performing comparably to Gemini 3 Pro and OpenAI's GPT-5.2. This upgrade enhances Google's AI products with improved visual understanding, making advanced AI capabilities more accessible for everyday workflows and data analysis. Google has now launched Gemini 3 Flash, a faster and more cost-effective AI model based on Gemini 3. According to Google, Gemini 3 Flash is up to three times faster than Gemini 2.5 Flash, and it outperforms previous Flash models in all internal tests. In several benchmark tests, Gemini 3 Flash performed on par with both Gemini 3 Pro and OpenAI's GPT-5.2. In the multimodal test MMMU-Pro, it even topped the list with a result of 81.2 percent. The Flash model is supposed to be adapted for fast and repetitive workflows.

Stone, who is the founder and director of the Learning Agents Research Group at his university, has spent decades thinking about the possibility of AI agents. They are, he says, systems that "sense the environment, decide what to do and take an action". Put in those terms, it may feel as if AI agents have been with us for years. For instance, IBM's Deep Blue computer appeared to have reacted to events on a real-world chessboard to beat former World Chess Champion Garry Kasparov in 1997. But Deep Blue wasn't an agentic AI, says Stone. "It was decision-making, but it wasn't sensing or acting," he says.

This submission is concerned with unconstrained nonconvex stochastic optimization problems, in a setting in which the function to optimize is only available through stochastic estimates. Obtaining points satisfying second-order necessary optimality conditions has been a recent topic of interest at NIPS, as such points can be as good as global minima on several problems arising from machine learning. The authors present new complexity results that improve over the existing complexity bounds for finding an approximate local minimum, identified as a point for which the gradient norm is less than a threshold \epsilon and the minimum Hessian eigenvalue is at least -\sqrt{\epsilon} . By assuming that the objective function possesses a Lipschitz continous third-order derivative, the authors are able to guarantee a larger decrease for steps of negative curvature type: this argument is the key for obtaining lower terms in the final complexity bounds and, as a result, lower dependency on the tolerance \epsilon compared to other techniques ( \epsilon {-10/3} versus \epsilon {-7/2} in previous works). The authors conduct a thorough review of the related literature, and discuss the main differences between existing algorithms and theirs in Section 2. This literature review appears exhaustive, and identifies key differences between this work and the cited ones.

The team has developed an app called PupilScreen that uses video and a smartphone's camera flash to record and calculate how the pupils respond to light. Assessing head trauma due to, for example, sports injuries or a car crash is typically done with either a pupilometer -- rarely found outside of hospitals -- or a mix of subjective evaluations like balancing, repeating a list of words or visually examining a pupil's response with a flashlight. To create PupilScreen and provide an objective assessment of potential head trauma, the researchers used deep learning tools to train a neural network how to find the pupil of the eye and track how it responds to a flash of light over the course of three seconds. A smartphone camera records the three second video and the light is provided by the camera's flash.