The Archive

Compares strategies for learning cardiac motion priors in implicit neural representations to accelerate optimization.

Uses diffusion/flow models as geometric maps to enable gradient descent on learned data manifolds.

Studies emotion steering in text-to-speech models via activation steering on speech language models and flow-matching modules.

Meta is developing plans for a cloud infrastructure business, selling access to AI compute power and models. The move would pit it against the big cloud providers like Amazon Web Services, Google Cloud, and Microsoft Azure.

Measures instability in LLM persona-driven outputs for multiple-choice QA across performance, outcome, and question dimensions.

Proposes ILLUME+, a post-hoc explainability framework for cancer drug response prediction using multivariate gene interactions.

Develops post-training pruning methods for Diffusion Transformers that account for their unique weight distribution and skip-connections.

Proposes GMHF framework combining meta-learning with human feedback to guide synthetic data generation for domain adaptation.

Introduces Graph-PRefLexOR, a graph-native reasoning model fine-tuned with GRPO for traceable materials discovery hypothesis generation.

Presents MAGNET, a multi-agent narrative engine with persona-grounded characters, and ATLAS, a verification pipeline for long-form story generation.

Proposes Valdi, latent diffusion world models for fast Model Predictive Control that combines online training with diffusion-based dynamics.

Frontier AI models maintain capability leads on most metrics while smaller models may converge on validation loss as scaling continues.

Mixed-precision LLM quantization shows perplexity-based layer importance poorly predicts reasoning performance; combining task and general data improves post-quantization results.

New benchmark for span-level hallucination detection in RAG systems across code, tool output, structured documents, and natural language.

MultiSynt/MT releases 4.8 trillion tokens of open synthetic parallel pre-training data across 36 European languages via Tower+ and OPUS-MT translation.

DeWorldSG framework generates 3D semantic scene graphs from RGB-D sequences using probabilistic Gaussian distributions and depth-guided filtering.

Flat minima optimization applied to 3D Gaussian Splatting improves generalization on sparse-view novel view synthesis.

Theoretical proof that binary tree mechanism achieves optimal log^(3/2) n error dependence for differentially private continual counting.

Study of ethos and pathos rhetorical appeals in silent readers' interpretations of social media messages using labeled datasets.

SEA architecture enables safe self-modifying agents by freezing base models and gating adaptations through anytime-valid certificates against error budgets.

Production-scale clinical NLP study of inference-time gating with Llama-3.3 70B generator and MMed-Llama-3.1 70B verifier over 167K narratives shows pattern-memory filtering limitations.

The Google Home Speaker is Google’s first smart speaker in years. And it’s pretty! | Photo: Jennifer Patison Tuohy / The Verge Smart speakers have spent the past few years searching for a compelling second act. Beyond music, timers, and controlling your lights, they've struggled to justify taking up space on the kitchen counter. AI promised to change that. Amazon debuted its new hardware powered by a revamped Alexa last fall, and now it's finally Google's turn. The Google Home Speaker is the company's first new smart speaker in six years and its first "built for Gemini." After years of neglec...

Berkeley BAIR lab celebrates 2026 Ph.D. cohort with work spanning robotics, LLMs, vision, safety, and AI for science.

AI Engineer World's Fair coverage: agent loops, software factories, forward-deployed engineering, and open model adoption emerging as key themes.

Vinton Cerf, one of the creators of the protocols underlying the internet, will step down as Google's chief internet evangelist next week.

Latent Space newsletter teases upcoming model releases (Sonnet 5, Fable 5) with minimal detail; appears to be placeholder or speculative content.

Anthropic said it would begin restoring access to the Fable on July 1.

Wayve’s offering is part of a growing trend of AI startups using employee tenders as a strategic tool to attract and retain talent.

Sierra's Natalie Meurer discusses convergence of product and forward-deployed engineers in software development.

After weeks of negotiating with the Trump administration, Anthropic is finally going to be able to bring Claude Fable 5 back online. In a post on X, Anthropic said it plans to begin restoring access tomorrow. Anthropic: We've received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow, and will share an update soon. We're grateful to our users for their patience, and to everyone who worked with us on redeploying the models. In early June, Anthropic sidelined Fable 5 - its consumer-facing model built the same ...