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 ...
Harvard Business Review

Begin with Trust

The first step to becoming a genuinely empowering leader by Frances X. Frei and Anne Morriss On a spring afternoon in 2017, Travis Kalanick, then the CEO of Uber, walked into a conference room at the ...