Generative AI

A prompt-based approach to minimal-edit grammatical error correction reached an F0.5 score of 78.32 on the BEA-2019 test set, closing the gap to the fine-tuned single-model state of the art (Staruch et al., 2025) to 0.38…

Researchers posted arXiv:2609.10811v1, introducing "overpainting," an image editing operation that controls where an edit occurs via a trimap of white (must edit), gray (may edit), and black (must not edit) pixels. The m…

A new arXiv paper (2609.10863v1) identifies a duality between continuous and discrete flow matching, showing that projecting continuous convex-interpolant paths with one-hot targets through a position-wise argmax yields …

Researchers introduced AcFlow, an inference-time controller that steers text-to-image diffusion transformers (DiTs) by transporting intermediate layer image-token activations through a learned concept-conditioned velocit…

Researchers introduced CamPilot, a multi-agent framework that integrates cinematographic planning and camera-work control to generate more coherent, logically structured movies, trained via a GRPO-based learning paradigm…

A new arXiv paper (2609.11538v1) shows that Generative Marginalization Models (MaMs) are equivalent to Generative Flow Networks (GFlowNets), and introduces Particle GFlowNets, which extends MaMs' sampling strategy to non…

A web developer argues that AI-assisted programming requires a new set of coding primitives, proposing components, features, and prompts as the building blocks developers should think in as the stack moves higher. The de…

A developer affiliated with We0.ai argues that AI-generated websites require a post-launch maintenance loop covering SEO audits, analytics, content updates, and task tracking, rather than treating the site as a one-time …

Microsoft has patented a multi-agent AI system that infers a user's unstated preferences before executing a task, according to a filing tracked by Patentlyze. The system routes a user's request and metadata to an input a…

Datasette released versions 1.0a39 and 0.65.4 to patch security flaws found in an audit run with Claude Fable 5.1, GPT-5.6, and GPT-6 Astra, according to the project's release notes. The audit was prompted by issues repo…

A developer who switched from Anthropic's Claude Code to OpenAI's Codex (Terra/Sol/Astra) reported that Codex's Astra model generated a one-line git commit message — "Fix YouTube resolver playback and isolate proxy traff…

A paper submitted to arXiv on 9 Sep 2026 argues that the AI text watermarks now required under Article 50 of the EU AI Act, which took effect on August 2, 2026, cannot currently be verified by users or regulators. Anthro…

Researchers from Duke University and Tsinghua University published a paper on September 3, 2026 introducing PlaidQ, a continuous Gaussian latent-diffusion language model for code generation. Distilling the diffusion proc…

Extreme Networks Inc. made its Agent ONE Coworker generally available Wednesday to all Extreme Platform ONE customers worldwide as part of their existing subscription, following 18 months of the company's own production …

OpenAI Group PBC launched ChatGPT for Financial Services, a version of ChatGPT built for financial analysis, modeling and client materials, powered by GPT-6 Astra and available only to "eligible institutions" that hold a…

California Governor Gavin Newsom signed bipartisan legislation that creates the nation's strongest companion chatbot regulations and bans addictive social media features such as autoplay and algorithmic feeds based on us…

Obra launched a project-finance tool that turns receipts sent via WhatsApp, email, or photo into live project numbers, with AI extracting amount, vendor, date, category, and tax for one-tap human confirmation. The compan…

YuE2 (best-of-8) scored 6.9632 on SongBench, the highest observed mean among 15 evaluated settings on WildSongBench's 192 prompts, according to a September 5, 2026 evaluation, edging out Suno v5 at 6.8721. The same resul…

Anthropic's new documentation for Claude Opus 5 and Fable 5.1 advises users to hand over a complete task specification up front rather than a step-by-step script, warning that skills and prompts written for older models …

Higsfield's Genjutsu model applies a single AI-generated stylized still image onto recorded video to produce animated-style footage that preserves the original motion, expressions, and audio, according to a practical wor…