Topics

A new small AI model with just 7 million parameters outperforms much larger ones on tough puzzles, suggesting that smart design can replace massive scale in solving complex problems.

MIT researchers craft a periodic table to connect classical machine learning algorithms, revealing new ways to improve AI models.

Correlational Elegant Normal Form integrates learned variable correlations into symbolic normalization, restoring efficient SAT search and evolutionary modeling while preserving canonical structure in highly structured domains.

Fewer good examples train AI for complex reasoning tasks with impressive accuracy across diverse benchmarks.

KAIST's SHARE method allows quick creation of virtual spaces using ordinary images, bypassing the need for precise measurements or many photos.

New AI agent AlphaEvolve uses large language models to discover and optimize algorithms, improving efficiency in computing and mathematics.

Researchers investigate how hidden network shapes affect brain activity, climate, and AI in a study of topological dynamics.

New quantum algorithm runs thousands of times faster than supercomputers, advancing molecular studies.

A new AI method could overcome some important limitations of the transformer architecture and help develop LLMs with long term memory.

Helping protect spacecraft from colliding with objects like asteroids and satellites