Victor Carbune

Yu-Chung Hsiao, Fedir Zubach, Gilles Baechler et al. · deepmind

We introduce ScreenQA, a novel benchmarking dataset designed to advance screen content understanding through question answering. The existing screen datasets are focused either on low-level structural and component understanding, or on a much higher-level composite task such as navigation and task completion for autonomous agents. ScreenQA attempts to bridge this gap. By annotating 86k question-answer pairs over the RICO dataset, we aim to benchmark the screen reading comprehension capacity, thereby laying the foundation for vision-based automation over screenshots. Our annotations encompass full answers, short answer phrases, and corresponding UI contents with bounding boxes, enabling four subtasks to address various application scenarios. We evaluate the dataset's efficacy using both open-weight and proprietary models in zero-shot, fine-tuned, and transfer learning settings. We further demonstrate positive transfer to web applications, highlighting its potential beyond mobile applications.

Ondrej Skopek, Rahul Aralikatte, Sian Gooding et al.

Despite recent advances, evaluating how well large language models (LLMs) follow user instructions remains an open problem. While evaluation methods of language models have seen a rise in prompt-based approaches, limited work on the correctness of these methods has been conducted. In this work, we perform a meta-evaluation of a variety of metrics to quantify how accurately they measure the instruction-following abilities of LLMs. Our investigation is performed on grounded query-based summarization by collecting a new short-form, real-world dataset riSum, containing 300 document-instruction pairs with 3 answers each. All 900 answers are rated by 3 human annotators. Using riSum, we analyze the agreement between evaluation methods and human judgment. Finally, we propose new LLM-based reference-free evaluation methods that improve upon established baselines and perform on par with costly reference-based metrics that require high-quality summaries.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Victor Carbune, Hassan Mansoor, Fangyu Liu et al.

Vision-language models (VLMs) are achieving increasingly strong performance on multimodal tasks. However, reasoning capabilities remain limited particularly for smaller VLMs, while those of large-language models (LLMs) have seen numerous improvements. We propose a technique to transfer capabilities from LLMs to VLMs. On the recently introduced ChartQA, our method obtains state-of-the-art performance when applied on the PaLI3-5B VLM by \citet{chen2023pali3}, while also enabling much better performance on PlotQA and FigureQA. We first improve the chart representation by continuing the pre-training stage using an improved version of the chart-to-table translation task by \citet{liu2023deplot}. We then propose constructing a 20x larger dataset than the original training set. To improve general reasoning capabilities and improve numerical operations, we synthesize reasoning traces using the table representation of charts. Lastly, our model is fine-tuned using the multitask loss introduced by \citet{hsieh2023distilling}. Our variant ChartPaLI-5B outperforms even 10x larger models such as PaLIX-55B without using an upstream OCR system, while keeping inference time constant compared to the PaLI3-5B baseline. When rationales are further refined with a simple program-of-thought prompt \cite{chen2023program}, our model outperforms the recently introduced Gemini Ultra and GPT-4V.

Harrison Lee, Samrat Phatale, Hassan Mansoor et al.

Reinforcement learning from human feedback (RLHF) has proven effective in aligning large language models (LLMs) with human preferences, but gathering high-quality preference labels is expensive. RL from AI Feedback (RLAIF), introduced in Bai et al., offers a promising alternative that trains the reward model (RM) on preferences generated by an off-the-shelf LLM. Across the tasks of summarization, helpful dialogue generation, and harmless dialogue generation, we show that RLAIF achieves comparable performance to RLHF. Furthermore, we take a step towards "self-improvement" by demonstrating that RLAIF can outperform a supervised fine-tuned baseline even when the AI labeler is the same size as the policy, or even the exact same checkpoint as the initial policy. Finally, we introduce direct-RLAIF (d-RLAIF) - a technique that circumvents RM training by obtaining rewards directly from an off-the-shelf LLM during RL, which achieves superior performance to canonical RLAIF. Our results suggest that RLAIF can achieve performance on-par with using human feedback, offering a potential solution to the scalability limitations of RLHF.

Victor Carbune, Pedro Gonnet, Thomas Deselaers et al.

We describe an online handwriting system that is able to support 102 languages using a deep neural network architecture. This new system has completely replaced our previous Segment-and-Decode-based system and reduced the error rate by 20%-40% relative for most languages. Further, we report new state-of-the-art results on IAM-OnDB for both the open and closed dataset setting. The system combines methods from sequence recognition with a new input encoding using Bézier curves. This leads to up to 10x faster recognition times compared to our previous system. Through a series of experiments we determine the optimal configuration of our models and report the results of our setup on a number of additional public datasets.

Victor Carbune, Thierry Coppey, Alexander Daryin et al.

We present SmartChoices, an approach to making machine learning (ML) a first class citizen in programming languages which we see as one way to lower the entrance cost to applying ML to problems in new domains. There is a growing divide in approaches to building systems: on the one hand, programming leverages human experts to define a system while on the other hand behavior is learned from data in machine learning. We propose to hybridize these two by providing a 3-call API which we expose through an object called SmartChoice. We describe the SmartChoices-interface, how it can be used in programming with minimal code changes, and demonstrate that it is an easy to use but still powerful tool by demonstrating improvements over not using ML at all on three algorithmic problems: binary search, QuickSort, and caches. In these three examples, we replace the commonly used heuristics with an ML model entirely encapsulated within a SmartChoice and thus requiring minimal code changes. As opposed to previous work applying ML to algorithmic problems, our proposed approach does not require to drop existing implementations but seamlessly integrates into the standard software development workflow and gives full control to the software developer over how ML methods are applied. Our implementation relies on standard Reinforcement Learning (RL) methods. To learn faster, we use the heuristic function, which they are replacing, as an initial function. We show how this initial function can be used to speed up and stabilize learning while providing a safety net that prevents performance to become substantially worse -- allowing for a safe deployment in critical applications in real life.