Stability AI Introduces StableChat: A Research Preview of Conversational AI Assistant Similar to ChatGPT or Claude

Stability AI has introduced Stable Chat, a platform that closely mirrors the familiar format of ChatGPT. However, upon initial evaluation, there are areas in which this offering could benefit from further refinement and enhancement.

The interface provided by Stable Chat allows users to engage in text-based conversations, bearing a striking resemblance to the format of ChatGPT. Despite this similarity, the platform does exhibit limitations. Notably, it needs a chat history feature and the ability to establish separate chat sessions. A notable inconvenience arises from conversation data being lost upon refreshing the interface.

At the core of this innovative chat interface lies the recently released Stable Beluga model, a creation of Stability AI. Initial feedback from online commentators suggests that while the model can understand Russian words, its proficiency leans towards recognizing individual words rather than constructing coherent language structures.

Interactions with the Beluga model have unveiled several key observations:

The model adheres to stringent safety protocols, demonstrating a degree of caution in generating specific types of content. For example, it has been observed to decline to generate content of an erotic nature initially.

The model’s content generation displays diversity, offering varied responses when presented with identical prompts. This divergence sometimes extends to altering content formats based on its discretion. This encompasses structuring information as lists, crafting paragraphs with explanatory content, or formulating sections featuring lists of words.

Regarding language style, the model occasionally needs to be more consistent. Instances have been noted where the tone and structure of generated content appear to shift, indicating areas that could be improved.

Unique to the model’s responses is the occasional inclusion of phrases that introduce an element of unpredictability. For instance, in response to queries related to board games, the model produced unique expressions such as “Peppermint drinker,” “Medusonki,” and “Lamb cub.”

Curiously, the model sometimes refers users to external resources. For instance, it has suggested using Google Translator to expand word options for specific topics.

One notable concern is the model’s stability and consistency. Users have reported occasional disconnects and instances of short timeouts for inactivity within the chat window.

Compared to the well-established ChatGPT, it becomes evident that there are areas within Stable Chat that would benefit from further refinement. While Stability AI’s research preview showcases promise and offers distinctive content generation, it does come with occasional inconsistencies.

Check out the Reference Article and Try it now. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our 28k+ ML SubReddit, 40k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.