The broad set of deep generative models (DGMs) has achieved remarkable advances. However, it is often difficult to incorporate rich structured domain knowledge with the end-to-end DGMs. Posterior regularization (PR) offers a principled framework to impose structured constraints on probabilistic models, but has limited applicability to the diverse DGMs that can lack a Bayesian formulation or even explicit density evaluation. PR also requires constraints to be fully specified {\it a priori}, which is impractical or suboptimal for complex knowledge with learnable uncertain parts. In this paper, we establish mathematical correspondence between PR and reinforcement learning (RL), and, based on the connection, expand PR to learn constraints as the extrinsic reward in RL. The resulting algorithm is model-agnostic to apply to any DGMs, and is flexible to adapt arbitrary constraints with the model jointly. Experiments on human image generation and templated sentence generation show models with learned knowledge constraints by our algorithm greatly improve over base generative models.

We address the problem of learning semantic representation of questions to measure similarity between pairs as a continuous distance metric. Our work naturally extends Word Mover's Distance (WMD) [1] by representing text documents as normal distributions instead of bags of embedded words. Our learned metric measures the dissimilarity between two questions as the minimum amount of distance the intent (hidden representation) of one question needs to travel to match the intent of another question. We first learn to repeat, reformulate questions to infer intents as normal distributions with a deep generative model [2] (variational auto encoder). Semantic similarity between pairs is then learned discriminatively as an optimal transport distance metric (Wasserstein 2) with our novel variational siamese framework. Among known models that can read sentences individually, our proposed framework achieves competitive results on Quora duplicate questions dataset. Our work sheds light on how deep generative models can approximate distributions (semantic representations) to effectively measure semantic similarity with meaningful distance metrics from Information Theory.

In high dimensional settings, density estimation algorithms rely crucially on their inductive bias. Despite recent empirical success, the inductive bias of deep generative models is not well understood. In this paper we propose a framework to systematically investigate bias and generalization in deep generative models of images by probing the learning algorithm with carefully designed training datasets. By measuring properties of the learned distribution, we are able to find interesting patterns of generalization. We verify that these patterns are consistent across datasets, common models and architectures.

We consider the semi-supervised clustering problem where crowdsourcing provides noisy information about the pairwise comparisons on a small subset of data, i.e., whether a sample pair is in the same cluster. We propose a new approach that includes a deep generative model (DGM) to characterize low-level features of the data, and a statistical relational model for noisy pairwise annotations on its subset. The two parts share the latent variables. To make the model automatically trade-off between its complexity and fitting data, we also develop its fully Bayesian variant. The challenge of inference is addressed by fast (natural-gradient) stochastic variational inference algorithms, where we effectively combine variational message passing for the relational part and amortized learning of the DGM under a unified framework. Empirical results on synthetic and real-world datasets show that our model outperforms previous crowdsourced clustering methods.

Where OpenAI's technology could show up in Iran Three places to watch, from the margins of war to the center of combat. It's been just over two weeks since OpenAI reached a controversial agreement to allow the Pentagon to use its AI in classified environments. There are still pressing questions about what exactly OpenAI's agreement allows for; Sam Altman said the military can't use his company's technology to build autonomous weapons, but the agreement really just demands that the military follow its own (quite permissive) guidelines about such weapons. OpenAI's other main claim, that the agreement will prevent use of its technology for domestic surveillance, appears equally dubious . It's not the first tech giant to embrace military contracts it had once vowed never to enter into, but the speed of the pivot was notable. Perhaps it's just about money; OpenAI is spending lots on AI training and is on the hunt for more revenue (from sources including ads).

The encyclopedia company's lawsuit also said ChatGPT cannibalizes traffic to the Britannica and Merriam-Webster websites. OpenAI has been hit with another lawsuit. According to the lawsuit, ChatGPT generates made-up content or ' hallucinations ' and falsely attributes them to Encyclopedia Britannica. The lawsuit doesn't specify an amount for monetary damages, but Britannica is also seeking an injunction to prevent OpenAI from repeating these accusations. When reached out for comment, a spokesperson for OpenAI told Engadget that, ChatGPT helps enhance human creativity, advance scientific discovery and medical research, and enable hundreds of millions of people to improve their daily lives.

OpenAI's adult mode reportedly won't generate pornographic audio, images or video The company's own council on wellbeing and AI appears to be against the feature. OpenAI's forthcoming adult mode will allow users to engage in lewd conversations with ChatGPT, but not use the chatbot to generate explicit images, audio or video. In response to reporting from an OpenAI spokesperson characterized the upcoming release as capable of producing smut rather than pornography. OpenAI CEO Sam Altman first floated the idea of allowing people to use ChatGPT for erotica, saying the company wanted to treat adult users like adults. OpenAI originally planned to release adult mode at the start of 2026.

The Online Services Accord Against Scams was signed by major tech companies including Google, Microsoft and OpenAI. A coalition of Big Tech companies is working on a more comprehensive solution to combat online scams . As first reported by, Google, Microsoft, LinkedIn, Meta, Amazon, OpenAI, Adobe and Match Group announced the signing of the Online Services Accord Against Scams. The new agreement is meant to put up a united industry-wide front against online fraud and scams, particularly those from sophisticated criminal networks that use multiple platforms. According to the report, the measures will include adding fraud detection tools, introducing new user security features, and requiring more robust verification for financial transactions.

The company launched its Sora 2 model in September 2025 alongside a dedicated Sora app. OpenAI plans to add its Sora video generation model directly into ChatGPT, reports . The standalone Sora app was seen as a smash hit when it launched alongside Sora 2 in September 2025, but interest in the video generation app has fallen in the time since as users ran into limits on the amount and kinds of videos they could create. Adding Sora to the ChatGPT could give the model a second life, and ideally grow the ChatGPT app's weekly active users from the 900 million OpenAI reported in February, to a billion or more. According to, the standalone Sora app will stick around after the model is integrated, even though the app has fallen out of the App Store's top 100 free apps and only a small number of users reportedly share their videos publicly in the app.

PCWorld reports that OpenAI plans to integrate its Sora video generator directly into ChatGPT, making AI video creation more accessible to users. This integration could lead to changes in ChatGPT's subscription plans and pricing structure due to the high costs of running video-based generative AI.