EVI 2 Large (LLM improvements)
in progress
J
Jake Marlo
Currently, EVI 2 uses a small LLM that is less intelligent but can interact quickly in conversations. A larger Hume LLM model is in training, with improved logic, knowledge, and memory capabilities.
This will reduce a lot of issues, including:
- Content coherence issues - Instances of providing random or irrelevant information, especially in longer conversations
- Context retention problems - Difficulty in maintaining context from previous parts of the conversation, lack of memory
- Repetitions - EVI responding to itself, saying repetitive phrases, or repeating the same thing twice.
- Overly terse 1-sentence responses - EVI sometimes says something very brief or incomplete in 1 sentence - e.g. EVI will respond with one thing that says it complies with your request, but then it doesn't actually do the thing. This will be improved substantially with the larger voice-language model for EVI 2.
J
Jake Marlo
in progress