Matyas Bohacek
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Matyas Bohacek has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Saad Hassan
Department of Computer Science, Tulane University, New Orleans, Louisiana, USA
3 shared publications - Chaelin Kim
Tulane University, New Orleans, Louisiana, USA
2 shared publications - Denise Crochet
Tulane University, New Orleans, Louisiana, USA
2 shared publications Filip Trhlík
Faculty of Social Sciences Charles University Prague Czech Republic
2 shared publications- Hany Farid
University of California Berkeley Berkeley CA USA
2 shared publications - Ignacio Vilanova Echavarri
Stanford University Imperial College London
2 shared publications
Latest Publications
- Open Access
- 1 July 2026
- Published by Association for Research in Vision and Ophthalmology (ARVO) in Journal of Vision
- Vol. 26 (7) , 3
AI-generated faces are becoming more trustworthy.
The development of generative artificial intelligence (GAI) has democratized access to high-quality image generation, creating both opportunities and risks. Research shows that generative adversarial networks (GANs) can generate faces that are nearly indistinguishable from and more trustworthy than real ones. However, the literature currently contains no direct comparison of the realism and trustworthiness of GAN-synthesized faces and faces produced by more recent models. Here we test whether faces generated by diffusion models (DMs), a new and more powerful GAI architecture, elicit trust and have similarly passed through the uncanny valley. In Experiment 1, participants (N = 169) saw 96 facial images and for each face indicated whether it was real or AI generated. In Experiment 2, participants (N = 87) again saw 96 facial images and this time were asked to rate how trustworthy each face appeared. We find that DMs produce less photo-realistic faces than GANs, but those faces are rated as more trustworthy than GANs and real faces. As the realism and availability of GAI continue to increase, it is now-more than ever-critical to understand this threat and develop strategies to mitigate potential harms to individuals, organizations, and democracies.
- 1 July 2026
- Published by Association for Research in Vision and Ophthalmology (ARVO) in Journal of Vision
- Vol. 26 (7) , 3
Open AccessAI-generated faces are becoming more trustworthy.
The development of generative artificial intelligence (GAI) has democratized access to high-quality image generation, creating both opportunities and risks. Research shows that generative adversarial networks (GANs) can generate faces that are nearly indistinguishable from and more trustworthy than real ones. However, the literature currently contains no direct comparison of the realism and trustworthiness of GAN-synthesized faces and faces produced by more recent models. Here we test whether faces generated by diffusion models (DMs), a new and more powerful GAI architecture, elicit trust and have similarly passed through the uncanny valley. In Experiment 1, participants (N = 169) saw 96 facial images and for each face indicated whether it was real or AI generated. In Experiment 2, participants (N = 87) again saw 96 facial images and this time were asked to rate how trustworthy each face appeared. We find that DMs produce less photo-realistic faces than GANs, but those faces are rated as more trustworthy than GANs and real faces. As the realism and availability of GAI continue to increase, it is now-more than ever-critical to understand this threat and develop strategies to mitigate potential harms to individuals, organizations, and democracies.
- Open Access
- 25 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 840-872
Unsteady Metrics and Benchmarking Cultures of AI Model Builders
- 25 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 840-872
Open AccessUnsteady Metrics and Benchmarking Cultures of AI Model Builders
- Open Access
Compliance Rating Scheme: A Data Provenance Framework for Generative AI Datasets
Generative Artificial Intelligence (GAI) has experienced exponential growth in recent years, partly facilitated by the abundance of large-scale open-source datasets. These datasets are often built using unrestricted and opaque data collection practices. While most literature focuses on the development and applications of GAI models, the ethical and legal considerations surrounding the creation of these datasets are often neglected. In addition, as datasets are shared, edited, and further reproduced online, information about their origin, legitimacy, and safety often gets lost. To address this gap, we introduce the Compliance Rating Scheme (CRS), a framework designed to evaluate dataset compliance with critical transparency, accountability, and security principles. We also release an open-source Python library built around data provenance technology to implement this framework, allowing for seamless integration into existing dataset-processing and AI training pipelines. The library is simultaneously reactive and proactive, as in addition to evaluating the CRS of existing datasets, it equally informs responsible scraping and construction of new datasets.
Open AccessCompliance Rating Scheme: A Data Provenance Framework for Generative AI Datasets
Generative Artificial Intelligence (GAI) has experienced exponential growth in recent years, partly facilitated by the abundance of large-scale open-source datasets. These datasets are often built using unrestricted and opaque data collection practices. While most literature focuses on the development and applications of GAI models, the ethical and legal considerations surrounding the creation of these datasets are often neglected. In addition, as datasets are shared, edited, and further reproduced online, information about their origin, legitimacy, and safety often gets lost. To address this gap, we introduce the Compliance Rating Scheme (CRS), a framework designed to evaluate dataset compliance with critical transparency, accountability, and security principles. We also release an open-source Python library built around data provenance technology to implement this framework, allowing for seamless integration into existing dataset-processing and AI training pipelines. The library is simultaneously reactive and proactive, as in addition to evaluating the CRS of existing datasets, it equally informs responsible scraping and construction of new datasets.
- Open Access
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 12150-12159
Compliance Rating Scheme: A Data Provenance Framework for Generative AI Datasets
- 27 October 2025
- Published by Association for Computing Machinery (ACM)
- p. 12150-12159
Open AccessCompliance Rating Scheme: A Data Provenance Framework for Generative AI Datasets
- Open Access
- 28 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 80-94
Towards an AI-Driven Video-Based American Sign Language Dictionary: Exploring Design and Usage Experience with Learners
- 28 April 2025
- Published by Association for Computing Machinery (ACM)
- p. 80-94
Open AccessTowards an AI-Driven Video-Based American Sign Language Dictionary: Exploring Design and Usage Experience with Learners
- Open Access
- 27 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Pose-aware Large Language Model Interface for Providing Feedback to Sign Language Learners
Sign language learners often find it challenging to self-identify and correct mistakes, and so many turn to automated methods that provide sign language feedback. However, they find that existing methods either require specialized equipment or lack robustness. They, therefore, have to seek human tutors or give up on the inquiry altogether. To overcome the barriers in accessibility and robustness, we build a large language model (LLM)-based tool for that provide feedback to sign language learners. The tool can analyze videos from diverse camera and background settings without specialized equipment thanks to a sign language segmentation and keyframe identification model. Using a pose-aware LLM, the tool can then produce feedback in written language. We present our tool as a demo web application, opening its implementation into specialized learning applications.
- 27 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessPose-aware Large Language Model Interface for Providing Feedback to Sign Language Learners
Sign language learners often find it challenging to self-identify and correct mistakes, and so many turn to automated methods that provide sign language feedback. However, they find that existing methods either require specialized equipment or lack robustness. They, therefore, have to seek human tutors or give up on the inquiry altogether. To overcome the barriers in accessibility and robustness, we build a large language model (LLM)-based tool for that provide feedback to sign language learners. The tool can analyze videos from diverse camera and background settings without specialized equipment thanks to a sign language segmentation and keyframe identification model. Using a pose-aware LLM, the tool can then produce feedback in written language. We present our tool as a demo web application, opening its implementation into specialized learning applications.
- Open Access
- 23 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 51-55
Evaluation of Concept Erasing for Artistic Styles in Diffusion Models
Artists have been raising the alarm about unfair and opaque practices of generative AI models for nearly two years. In a technological attempt to answer these concerns, concept erasing has promised to give artists more agency in determining whether new generative AI models get to clone their signature artistic styles. The idea behind this method is simple: take an already trained AI model, choose an artist’s style to remove from the model, and apply minimal changes to suppress that concept. Simple integration into existing pipelines makes it a particularly appealing solution. However, the efficacy of this method has so far been examined for only one artist’s style at a time. This paper, thus, examines concept erasing at scale, simulating the hypothetical application of this technology for simultaneously erasing many artists. It finds that the method does not reliably erase the concepts in cases with more than one artist’s style, highlighting a gap in the technology and the literature around it.
- 23 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 51-55
Open AccessEvaluation of Concept Erasing for Artistic Styles in Diffusion Models
Artists have been raising the alarm about unfair and opaque practices of generative AI models for nearly two years. In a technological attempt to answer these concerns, concept erasing has promised to give artists more agency in determining whether new generative AI models get to clone their signature artistic styles. The idea behind this method is simple: take an already trained AI model, choose an artist’s style to remove from the model, and apply minimal changes to suppress that concept. Simple integration into existing pipelines makes it a particularly appealing solution. However, the efficacy of this method has so far been examined for only one artist’s style at a time. This paper, thus, examines concept erasing at scale, simulating the hypothetical application of this technology for simultaneously erasing many artists. It finds that the method does not reliably erase the concepts in cases with more than one artist’s style, highlighting a gap in the technology and the literature around it.
- Open Access
- 10 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 2
From Open-Source to Primetime: The Making of an AI News Anchor and its Role in the New Landscape of Disinformation
- 10 June 2024
- Published by Association for Computing Machinery (ACM)
- p. 2
Open AccessFrom Open-Source to Primetime: The Making of an AI News Anchor and its Role in the New Landscape of Disinformation
- Open Access
- 22 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Sign Spotter: Design and Initial Evaluation of an Automatic Video-Based American Sign Language Dictionary System
- 22 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessSign Spotter: Design and Initial Evaluation of an Automatic Video-Based American Sign Language Dictionary System