Previous LLMs operated in stateless isolation. Each chat session began with amnesia. AIRevolution -v0.3.5- -Akaime- changes this through Persistent Episodic Memory, a 350MB on-disk vector database that retains compressed embeddings of all user interactions across sessions.
How it works:
Example: If you asked the model two weeks ago to help outline a sci-fi novel about terraforming Mars, and today you ask “What’s a good subversion of the lone astronaut trope?”, Akaime will surface that earlier discussion without being prompted. The result is a conversation that feels less like talking to a search engine and more like speaking with a research assistant who remembers your pet theories. AIRevolution -v0.3.5- -Akaime-
Most models waste FLOPS on simple questions and struggle with complex ones, because they use the same inference budget per token. AIRevolution -v0.3.5- introduces a complexity predictor (a 3-layer transformer trained on 2 million query-response pairs) that: Previous LLMs operated in stateless isolation
The result: average response latency for simple factual questions dropped from 1.2s to 0.45s, while performance on the MMLU-Pro benchmark (complex reasoning) rose from 68.3% to 74.9%. Example: If you asked the model two weeks
Based on the version number (v0.3.5) and the creator (Akaime), this review focuses on the Minecraft Bedrock Edition Add-on. This mod is widely considered one of the most ambitious and polished "Mob Battle/AI" addons currently available.
Here is a detailed review of AIRevolution v0.3.5.