recently about their approach to solving certain problems which they say can be solved faster and with less energy consumption than Large Language Models . Their approach makes heavy use of exact probabilistic inference. PiLogic says that their inference engine is the most advanced in the world as benchmarked against Join Tree and other leading methods.
This approach doesn’t require huge data sets and specialized expensive hardware such as Graphics Processing Units . It has particular value for engineering use cases, doesn’t have hallucinations and gives results which are precise and accurate.
In addition to efficiency, ACs have other advantages. For example, it is possible to know precisely how much time and space is required to answer queries, and so the approach works well in the context of real-time requirements. Moreover, the AC can be embedded in many products and applications since it doesn’t require specialized hardware. These efficiency improvements also lead to energy savings for the entire inference process on an ongoing basis for end users.
PiLogic says that it has found a way to break this exponential growth in calculation complexity for many problems. They do this by using structure in the problem, particularly local structure. This may be zeros or repeated values in the model that can simplify the calculations needed. As a consequence, PiLogic says that if there is sufficient local structure, they can solve problems with treewidth into the 100’s, as shown above.