Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
eLife

Dynamics of mesoscale brain network during visual discrimination learning revealed by chronic, large-scale single-unit recording

During the acquisition of correct rejection response, rankings of functional connection separated for cortical and subcortical regions, which is predictive of the peak timing of visual information ...
The Hearty Soul on MSN

Scientists reveal first generation ever to be less intelligent than their parents

Historically, intelligence tests and large-scale assessments have shown that each new generation tends to outperform its ...

Guide Labs debuts a new kind of interpretable LLM

The company open-sourced an 8 billion parameter LLM, Steerling-8B, trained with a new architecture designed to make its ...