Kona is an energy-based AI model that solves by constraint solving, not individual token commitment. This reduces the compute requirement by 2-3 orders of magnitude, making running AI much less energy intensive, less harmful to the environment (less power = less heat = less cooling needed), as well as much less expensive. Much greater accuracy is a bonus 😉.

Since all situations requiring a solution can be defined by constraints, this is a promising avenue of AI research; one that might just make AI both useful and practical.

https://logicalintelligence.com/blog/energy-based-model-sudoku-demo

*The Sudoku demo is a limited application, but the results show the vast gulf between energy-based systems and traditional LLMs.

#AI #LogicalIntelligence #Kona

EBM vs. LLMs: Our Kona EBM a 96% vs. 2% Sudoku Benchmark

Our Energy-Based Model, Kona, solves 96.2% of Sudoku puzzles in ~313ms, while LLMs fail 98% of the time. See the demo and why this matters for industrial AI.

Kiva ilma pyöräillä, vaik vähän välist tuulikin turhan kovasti

#pyöräily #fillari #kona