Updates for Tuesday, March 31, 2026

ChatGPT is introducing new features for file storage and a personal library, allowing users to upload, store, and manage their own files directly within the platform. This significantly enhances its ability to reference and interact with user-provided data across sessions, making it a more powerful tool for personalized tasks and information management.

Runway is launching a $10 million fund and a "Builders program" to support early-stage AI startups. This initiative aims to back companies that are developing interactive, real-time "video intelligence" applications using Runway's AI video models. The move is designed to foster an ecosystem around Runway's technology and accelerate innovation in the AI video sector.

Google Maps is enhancing its features for over 350 Android Auto-compatible EV models, simplifying battery predictions and trip planning. This update aims to reduce range anxiety for electric vehicle owners by providing more accurate information and integrated tools for managing charging stops. It significantly improves the convenience and user experience for a large segment of the EV market.

This AI film generation engine, "forge-film," introduces a novel approach to accelerate content creation by modeling narrative dependencies between scenes as a Directed Acyclic Graph (DAG). It utilizes the Critical Path Method (CPM) for parallel scheduling, enabling simultaneous generation of film scenes. This method significantly speeds up the production process by leveraging scene independence, claiming to be the only engine to use this DAG/CPM-driven parallel scheduling for film.

fastrad is a new GPU-native radiomics library built on PyTorch, offering a 25x speedup over the CPU-only PyRadiomics, the current standard. It implements all 8 IBSI-compliant feature classes as native tensor operations, significantly accelerating radiomic feature extraction. This development addresses a critical bottleneck for large-scale medical image analysis.

A developer successfully trained a family of language models (BULaMU) for Luganda, a low-resource language, entirely from scratch. These models, ranging from 20M to 110M parameters, are highly compute-efficient and can run fully offline on Android devices without needing a GPU or internet connection. An accompanying Android app, E.A.S.T., allows users to interact with these on-device AI capabilities, significantly expanding access to AI for underserved communities.

A PhD student developed "rnn.news," a personal research newspaper designed to filter the overwhelming volume of arXiv pre-prints. Users can email their research interests and receive a weekly, journalistic-style digest of highly relevant papers. This tool aims to help researchers efficiently discover crucial insights amidst the noise.

OpenAI has introduced workspace analytics for its ChatGPT Enterprise and Education versions. This new feature allows organizations to monitor usage and gain insights into how their teams or students are interacting with ChatGPT, enhancing manageability and demonstrating value for large-scale deployments.

This paper introduces Metric Similarity Analysis (MSA), a novel method that uses Riemannian geometry to compare the intrinsic geometry of neural network representations. Unlike existing methods that focus on extrinsic geometry, MSA aims to capture subtle yet crucial distinctions between fundamentally different neural network solutions. This approach promises a deeper understanding of how neural networks solve tasks by analyzing their representational geometries more accurately.

This paper addresses the significant lack of variety and 'typicality bias' in modern Text-to-Image (T2I) diffusion models, which often produce a narrow set of visual solutions for a given prompt. It introduces a novel method called 'On-the-fly Repulsion in the Contextual Space' to enrich diversity in Diffusion Transformers. This approach aims to overcome the costly optimization typically required to incorporate feedback for generating a wider range of creative outcomes.

New research suggests recurrent networks can adapt online more efficiently without complex Jacobian propagation. This is achieved by leveraging temporal credit in the forward pass and using immediate derivatives, while also avoiding stale trace memory and normalizing gradient scales. An architectural rule predicts when gradient normalization is crucial, potentially simplifying online learning across various RNN architectures.

An "Ask HN" post describes a user's observation that a previously authentic website has rapidly become saturated with AI-generated content, now making up at least half of its articles. Despite the clear presence of "LLM-isms," user comments engage with the material without acknowledging its inauthenticity, prompting the author to question if resistance to AI-generated "slop" is diminishing.

Weave is a new AI tool designed to analyze and evaluate employees' AI coding usage and effectiveness, aiming to help companies optimize their spending on expensive AI tokens. However, the tool provides almost no methodology or insight into how its metrics are calculated, beyond stating it uses AI, likely LLMs analyzing other LLMs. This lack of transparency raises significant concerns about the fairness and accuracy of its evaluations.

A 17-year-old self-taught security researcher, Dhanush, discovered vulnerabilities in a multi-million ARR AI company, "B," which specializes in "neural phase locking." Using Burp Suite, he identified issues in the company's service communications, highlighting potential security flaws in a premium-only AI platform. His findings, shared for educational purposes, underscore the importance of thorough security testing even in established tech companies.

AI gateway startup LiteLLM has ended its partnership with Delve, the company that provided its security compliance certifications. This decision follows a recent and severe credential-stealing malware attack suffered by LiteLLM, highlighting significant supply chain security vulnerabilities within the AI ecosystem.

Weejur is a new tool designed to simplify website creation by allowing users to paste HTML and publish it for free using GitHub Pages. It aims to make website development accessible to non-technical individuals by leveraging AI/LLMs to streamline the process. This initiative supports the "small web" movement, encouraging more people to easily create and host their own online content.

The new Dean of Studies at La Plateforme, a tuition-free French CS school, is seeking community input to shape the future curriculum. The school trains 200 students annually from scratch to Bachelor's and Master's degrees in fields like software development, DevOps, AI/data, cybersecurity, and immersive systems, without prior degree requirements. This initiative aims to ensure their education remains relevant and effective for future tech professionals.

Muon, a training technique, has been rapidly adopted for Large Language Model (LLM) training using Transformers, yet its application in other contexts like Convolutional Neural Networks (ConvNets) is surprisingly limited. Despite initial reports of achieving training speed records on datasets like Cifar-10, its absence elsewhere raises questions about its scalability and broader applicability. The discussion seeks to understand why Muon's use seems confined to Transformers.

A Reddit user is seeking advice regarding the 2026 Google PhD Fellowship Program application. As a first-year PhD student, they are concerned about their lack of recent top publications and are asking for insights into the selection criteria, the importance of publications, and the 200-word project summary. They are looking for constructive answers from those familiar with the program.