Learn about the new features and enhancements included in Kore.ai Experience Optimization Platform v10.2, released on April 27, 2024.
Key features and enhancements included in this release are summarized below.
LLM and Generative AI
Custom LLM Integration Support for Rephrase Dialog Responses
Rephrase Dialog Responses now supports Custom LLMs in addition to commercial LLMs. This allows platform users to use the rephrasing feature with their own custom-trained language models and create customized prompts tailored to their specific use cases, models, and linguistic contexts, providing greater flexibility and control over the rephrasing process and conversational experiences. Learn more.
NLP
New Pre-trained MPNet Few-shot Model for Intent Detection
The few-shot model now supports the Pre-trained MPNet embedding model for intent detection. The Pre-trained MPNet model is the advanced version of MPNet. It has been pre-trained and fine-tuned for superior accuracy and precision in intent identification compared to MPNet. Learn more.
Advantages of using the Pre-trained MPNet Model:
- Understanding Negations: Pre-trained models get negations like “not” or “don’t” easily, unlike MPNet, which might mix things up. So, when you say, “I want to transfer funds,” and “I do not want to transfer funds,” pre-trained models know the difference, making them better at understanding what you really mean.
- Clear Intent Differentiation: Pre-trained models are great at telling apart similar things, which can be tricky for MPNet. For example, if you say, “I want to unblock a card,” or “I want to block a card,” pre-trained models can confidently tell the difference between wanting to block or unblock a card, unlike MPNet, which might get confused.
- Staying on Topic: Pre-trained models are good at sticking to the topic at hand. So, if you talk about increasing your credit card limit, they won’t start suggesting things unrelated to credit cards, unlike MPNet, which might go off track.
- Easier Training: Pre-trained models need less training data compared to MPNet. This means the model can learn the same things with fewer examples, making the training process faster and simpler.
Channels
Support for Thread Handling for Virtual Assistants in Slack Channels
The Platform now offers native support for threaded conversations in the Slack channel. Users can initiate a new thread from any message within a Slack channel or direct message group.
Additionally, the platform provides extended functionality for developers. It can automatically create a new thread whenever a user @mentions the virtual assistant in a Slack channel. This behavior is configurable, giving developers control over this feature.
Digital Forms
- Enable the “Off-the-Record Information” Flag for Digital Forms: On a digital form, when the field’s “Off the record” flag is enabled, the field data is cleared at the end of the user session and not stored in databases or logs.
- Digital Forms Date Picker Supports Japanese: The digital form’s Date Picker now supports the Japanese language if the bot language is Japanese.