A pioneering achievement in the field of artificial intelligence, the Nyx persistent memory layer has been successfully developed to significantly boost local AI performance. Recent benchmark tests have demonstrated a substantial improvement in output quality and context retention, with useful output quality increasing by 10 times and context retention by 7 times.

A comprehensive civic investigation task was conducted to test the capabilities of Nyx, involving the construction of a full politician profile and follow-up questions that required the recall of previously established details. The results showed a marked difference between the test with Nyx active and the cold start, with Nyx achieving 37.70 eTPS and 0.950 Continuity, compared to 3.87 eTPS and 0.138 Continuity without it.

The implications of this innovation are far-reaching, with significant potential for applications such as CivicLens, an evidence-first politician research tool. With the integration of Nyx, the AI can retain the entirety of a case, enabling a single, coherent investigation rather than disconnected queries.

Although still in its early stages, the core memory layer has exhibited promising results, with the numbers substantiating its potential. The development of Nyx raises an important question regarding the benchmarking of AI systems’ ability to complete tasks across multiple sessions, highlighting the need for further evaluation and development in this area.

Photo by Arkhi Tektōn on Pexels
Photos provided by Pexels