Chinmay Kulkarni

Michael Terry, Chinmay Kulkarni, Martin Wattenberg et al. · deepmind

Modern AI enables a high-level, declarative form of interaction: Users describe the intended outcome they wish an AI to produce, but do not actually create the outcome themselves. In contrast, in traditional user interfaces, users invoke specific operations to create the desired outcome. This paper revisits the basic input-output interaction cycle in light of this declarative style of interaction, and connects concepts in AI alignment to define three objectives for interactive alignment of AI: specification alignment (aligning on what to do), process alignment (aligning on how to do it), and evaluation alignment (assisting users in verifying and understanding what was produced). Using existing systems as examples, we show how these user-centered views of AI alignment can be used descriptively, prescriptively, and as an evaluative aid.

Meredith Ringel Morris, Carrie J. Cai, Jess Holbrook et al.

Card et al.'s classic paper "The Design Space of Input Devices" established the value of design spaces as a tool for HCI analysis and invention. We posit that developing design spaces for emerging pre-trained, generative AI models is necessary for supporting their integration into human-centered systems and practices. We explore what it means to develop an AI model design space by proposing two design spaces relating to generative AI models: the first considers how HCI can impact generative models (i.e., interfaces for models) and the second considers how generative models can impact HCI (i.e., models as an HCI prototyping material).

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.

Miguel AP Oliveira, Stephane Manara, Bruno Molé et al.

Individuals and organizations cope with an always-growing amount of data, which is heterogeneous in its contents and formats. An adequate data management process yielding data quality and control over its lifecycle is a prerequisite to getting value out of this data and minimizing inherent risks related to multiple usages. Common data governance frameworks rely on people, policies, and processes that fall short of the overwhelming complexity of data. Yet, harnessing this complexity is necessary to achieve high-quality standards. The latter will condition any downstream data usage outcome, including generative artificial intelligence trained on this data. In this paper, we report our concrete experience establishing a simple, cost-efficient framework that enables metadata-driven, agile and (semi-)automated data governance (i.e. Data Governance 4.0). We explain how we implement and use this framework to integrate 25 years of clinical study data at an enterprise scale in a fully productive environment. The framework encompasses both methodologies and technologies leveraging semantic web principles. We built a knowledge graph describing avatars of data assets in their business context, including governance principles. Multiple ontologies articulated by an enterprise upper ontology enable key governance actions such as FAIRification, lifecycle management, definition of roles and responsibilities, lineage across transformations and provenance from source systems. This metadata model is the keystone to data governance 4.0: a semi-automatised data management process that considers the business context in an agile manner to adapt governance constraints to each use case and dynamically tune it based on business changes.

Zijie J. Wang, Chinmay Kulkarni, Lauren Wilcox et al. · gatech

Prompt-based interfaces for Large Language Models (LLMs) have made prototyping and building AI-powered applications easier than ever before. However, identifying potential harms that may arise from AI applications remains a challenge, particularly during prompt-based prototyping. To address this, we present Farsight, a novel in situ interactive tool that helps people identify potential harms from the AI applications they are prototyping. Based on a user's prompt, Farsight highlights news articles about relevant AI incidents and allows users to explore and edit LLM-generated use cases, stakeholders, and harms. We report design insights from a co-design study with 10 AI prototypers and findings from a user study with 42 AI prototypers. After using Farsight, AI prototypers in our user study are better able to independently identify potential harms associated with a prompt and find our tool more useful and usable than existing resources. Their qualitative feedback also highlights that Farsight encourages them to focus on end-users and think beyond immediate harms. We discuss these findings and reflect on their implications for designing AI prototyping experiences that meaningfully engage with AI harms. Farsight is publicly accessible at: https://PAIR-code.github.io/farsight.

Sakharam Gawade, Shivam Akhouri, Chinmay Kulkarni et al.

Large Action Models (LAMs) have revolutionized intelligent automation, but their application in healthcare faces challenges due to privacy concerns, latency, and dependency on internet access. This report introduces an ondevice, multi-agent healthcare assistant that overcomes these limitations. The system utilizes smaller, task-specific agents to optimize resources, ensure scalability and high performance. Our proposed system acts as a one-stop solution for health care needs with features like appointment booking, health monitoring, medication reminders, and daily health reporting. Powered by the Qwen Code Instruct 2.5 7B model, the Planner and Caller Agents achieve an average RougeL score of 85.5 for planning and 96.5 for calling for our tasks while being lightweight for on-device deployment. This innovative approach combines the benefits of ondevice systems with multi-agent architectures, paving the way for user-centric healthcare solutions.

Pranav Khadpe, Chinmay Kulkarni, Geoff Kaufman

When members of ad-hoc virtual teams need to collectively ideate or deliberate, they often fail to engage with each others' perspectives in a constructive manner. At best, this leads to sub-optimal outcomes and, at worst, it can cause conflicts that lead to teams not wanting to continue working together. Prior work has attempted to facilitate constructive communication by highlighting problematic communication patterns and nudging teams to alter interaction norms. However, these approaches achieve limited success because they fail to acknowledge two social barriers: (1) it is hard to reset team norms mid-interaction, and (2) corrective nudges have limited utility unless team members believe it is safe to voice their opinion and that their opinion will be heard. This paper introduces Empathosphere, a chat-embedded intervention to mitigate these barriers and foster constructive communication in teams. To mitigate the first barrier, Empathosphere leverages the benefits of "experimental spaces" in dampening existing norms and creating a climate conducive to change. To mitigate the second barrier, Empathosphere harnesses the benefits of perspective-taking to cultivate a group climate that promotes a norm of members speaking up and engaging with each other. Empathosphere achieves this by orchestrating authentic socio-emotional exchanges designed to induce perspective-taking. A controlled study (N=110) compared Empathosphere to an alternate intervention strategy of prompting teams to reflect on their team experience. We found that Empathosphere led to higher work satisfaction, encouraged more open communication and feedback within teams, and boosted teams' desire to continue working together. This work demonstrates that ``experimental spaces,'' particularly those that integrate methods of encouraging perspective-taking, can be a powerful means of improving communication in virtual teams.

Sara Kingsley, Clara Wang, Alex Mikhalenko et al.

Digital platforms, including social networks, are major sources of economic information. Evidence suggests that digital platforms display different socioeconomic opportunities to demographic groups. Our work addresses this issue by presenting a methodology and software to audit digital platforms for bias and discrimination. To demonstrate, an audit of the Facebook platform and advertising network was conducted. Between October 2019 and May 2020, we collected 141,063 ads from the Facebook Ad Library API. Using machine learning classifiers, each ad was automatically labeled by the primary marketing category (housing, employment, credit, political, other). For each of the categories, we analyzed the distribution of the ad content by age group and gender. From the audit findings, we considered and present the limitations, needs, infrastructure and policies that would enable researchers to conduct more systematic audits in the future and advocate for why this work must be done. We also discuss how biased distributions impact what socioeconomic opportunities people have, especially when on digital platforms some demographic groups are disproportionately excluded from the population(s) that receive(s) content regulated by law.