recursive diffusion training converges to a smoothed limit
even with perfect learning, early stopping in diffusion models causes geometric drift toward a unique limiting distribution that acts as a low-pass filter on the data.
topic
even with perfect learning, early stopping in diffusion models causes geometric drift toward a unique limiting distribution that acts as a low-pass filter on the data.
a satellite ran a vision-language model in space for the first time, letting it answer natural language queries about earth imagery without human help.
a new method called lomc suppresses refusals in mixture-of-experts models using compact, support-gated edits that preserve general performance.
a new framework uses user portraits and nested rollout to let large language models dynamically tailor dialogue policies without offline training.
a study shows optimal transport with p-norm cost can recover rotation, translation, or scaling between source and target domains in 2d, enabling linear regression adaptation with scarce target data.
tech companies cite ai for record layoffs while insiders amass huge wealth, fueling skepticism and public anger.
a new framework extends decision trees to any bregman divergence loss, generalizing classic cart and enabling flexible, interpretable models for diverse data types.
a transformer-based scheduling policy trained with deep reinforcement learning solves open shop scheduling problems and scales to large instances without retraining.
llada.cpp is the first npu-aware framework for accelerating diffusion large language models on smartphones, using multi-block speculative decoding and other techniques.
ai speeds up coding but software engineering bottlenecks remain in deciding what to build, verifying outputs, and deep human understanding.
spacex's record ipo, anthropic's model shutdown, and india's ai sovereignty debate lead a week of market shifts and regulatory moves.
spacex's record ipo kicks off a rush of ai companies going public, with anthropic and openai close behind, creating ripple effects across industries.
india debates ai self-reliance after anthropic blocks models, meta unwinds manus deal under pressure, and state attorneys general investigate openai.
anthropic suspended access to its latest ai models for foreign nationals under a us directive, reigniting india's debate on technological dependence and domestic ai development.
simon willison releases luau-wasm 0.1a0, packaging the luau programming language as a webassembly wheel for use with pyodide in the browser.
simon willison explores methods to map sql query result columns back to their source table and column in sqlite, enabling richer data display in datasette.
pyodide now supports installing packages with webassembly wheels directly from pypi, removing the need for maintainers to manually host them.
meta begins operational separation from chinese ai startup manus after beijing's divestiture order on national security grounds.