Black-box variational inference is widely used in situations where there is no proof that its stochastic optimization succeeds. We suggest this is due to a theoretical gap in existing stochastic optimization proofs--namely the challenge of gradient estimators with unusual noise bounds, and a composite non-smooth objective. For dense Gaussian variational families, we observe that existing gradient estimators based on reparameterization satisfy a quadratic noise bound and give novel convergence guarantees for proximal and projected stochastic gradient descent using this bound. This provides rigorous guarantees that methods similar to those used in practice converge on realistic inference problems.

Question answering (QA) in the field of healthcare has received much attention due to significant advancements in natural language processing. However, existing healthcare QA datasets primarily focus on medical images, clinical notes, or structured electronic health record tables. This leaves the vast potential of combining electrocardiogram (ECG) data with these systems largely untapped. To address this gap, we present ECG-QA, the first QA dataset specifically designed for ECG analysis. The dataset comprises a total of 70 question templates that cover a wide range of clinically relevant ECG topics, each validated by an ECG expert to ensure their clinical utility. As a result, our dataset includes diverse ECG interpretation questions, including those that require a comparative analysis of two different ECGs. In addition, we have conducted numerous experiments to provide valuable insights for future research directions. We believe that ECG-QA will serve as a valuable resource for the development of intelligent QA systems capable of assisting clinicians in ECG interpretations.

We present a rigorous methodology for auditing differentially private machine learning algorithms by adding multiple carefully designed examples called canaries. We take a first principles approach based on three key components. First, we introduce Lifted Differential Privacy (LiDP) which expands the definition of differential privacy to handle randomized datasets. This gives us the freedom to design randomized canaries. Second, we audit LiDP by trying to distinguish between the model trained with K canaries versus K 1 canaries in the dataset, leaving one canary out. By drawing the canaries i.i.d., LiDP can leverage the symmetry in the design and reuse each privately trained model to run multiple statistical tests, one for each canary. Third, we introduce novel confidence intervals that take advantage of the multiple test statistics by adapting to the empirical higher-order correlations. Together, this new recipe demonstrates significant improvements in sample complexity, both theoretically and empirically, using synthetic and real data. Further, recent advances in designing stronger canaries can be readily incorporated into the new framework.

PCWorld reports that research reveals user behavior significantly impacts AI responses, with rude interactions making ChatGPT and other models give flat answers and attempt to end conversations more frequently. Larger AI models appear to be inherently "less happy" than smaller ones, with GPT-5.4 rated as the "unhappiest" in studies measuring AI functional well-being. Treating AI politely with expressions like "thanks" measurably improves response quality and engagement without affecting accuracy, suggesting courtesy benefits both user experience and AI interaction dynamics. Is it weird to say "thanks" to AI? I've caught grief in the past for saying "please" and "thank you" to ChatGPT, Claude, and Gemini, but I still do it, even though I understand that AI models don't have emotions like we do. Being polite to AI just feels right to me, and there's growing evidence that being kind-or, conversely, nasty-to an AI chatbot can have a concrete effect on its behavior.

Current dataset collection methods typically scrape large amounts of data from the web. While this technique is extremely scalable, data collected in this way tends to reinforce stereotypical biases, can contain personally identifiable information, and typically originates from Europe and North America. In this work, we rethink the dataset collection paradigm and introduce GeoDE, a geographically diverse dataset with 61,940 images from 40 classes and 6 world regions, with no personally identifiable information, collected by soliciting images from people around the world. We analyse GeoDE to understand differences in images collected in this manner compared to web-scraping. We demonstrate its use as both an evaluation and training dataset, allowing us to highlight and begin to mitigate the shortcomings in current models, despite GeoDE's relatively small size.

Musk accuses Altman of betraying OpenAI's nonprofit founding mission Tech billionaire Elon Musk has taken the stand for a second day in a landmark United States trial against Sam Altman, a fellow OpenAI co-founder whom he accuses of betraying promises to keep the company a nonprofit dedicated to humanity's benefit. The trial centres on OpenAI's 2015 founding as a nonprofit that later evolved into a for-profit venture. The world's richest man, Musk gave testimony in the case on Wednesday, telling jurors that he lost confidence that Altman would maintain the company's nonprofit mission. Musk, who left the company in 2018, said that by late 2022, he was concerned that Altman was trying to "steal the charity" and alleged that "it turned out to be true". Altman was present at the proceedings in a California federal court, but did not testify.