AI chatbots are evolving! Chaiverse connects developers and users for personalized AI experiences. Discover next-gen chat and support the democratization of AGI.
Many Chai users have been asking, “What is Chaiverse?” a feature currently obscurely advertised by the Chai developers. With a recent valuation of $450 million, the Chai AI team has been pouring their funding into the development of Chaiverse, with the aim of connecting world-class Large Language Model (LLM) developers directly to millions of Chat AI consumers, offering each user tailor-made combination of LLMs. This article provides a deep dive into the Chaiverse mechanism, how it works, and its implications for users.
Chaiverse – The Developer to Consumer Ecosystem
One major product that sets Chai AI apart from others is its developer-centric approach. Unlike other Generative AI products, which typically offer models developed in-house, Chai AI has recently introduced its model developer platform, Chaiverse. This platform enables the Large-Language-Model Developer community to train, submit, and test their models with real-world users in a consumer setting. According to the Chaiverse white paper, their mission is to accelerate the advent of AGI through massively distributed collaboration or crowdsourcing. Chai AI’s research team purports to be laser-focused on exploring what makes each Large Language Model (LLM) unique and improving person-to-LLM recommendations.
The operation of Chai AI’s ecosystem is quite straightforward. Developers can easily upload their language models using the chaiverse pip package. Once submitted, these models are optimized for rapid inference and hosted on a dedicated GPU cluster for enhanced efficiency. After the model is operational, users of the Chai App can engage with it through arena mode, providing developers immediate numerical and textual feedback.
This feedback, along with the ranking of a developer’s model on the public leaderboard, can be accessed through the chaiverse package. Cash prizes are awarded based on the developers’ standings in the competition, incentivizing them to strive for innovation and excellence in their model submissions.
How does Chaiverse affect Chai App Users?
While the Chai AI team is quite secretive about how it will be integrating Chaiverse into its consumer-facing Chai App, we have gained some insights into how this will be achieved through Chai’s white paper on Chaiverse. In Chai’s white paper, they claim to have been implementing an LLM controller, routing each message input (conversation tree) to an LLM that is well-placed at generating a response specifically for the given conversation tree.
This approach offers multiple advantages:
- Lower training costs: Training small-size models is accessible to a wider range of developers. Indeed, it is well known that training large-parameter LLMs is a prohibitively expensive task, and as a result is inaccessible to most machine learners.
- Lower inference costs: The inference cost of a multitude of small expert models is lower than that of a single large model.
- Higher iteration speed: Small models can be trained and deployed much faster than large ones, allowing for much shorter feedback loops and higher iteration speeds.
- Early results are promising: A carefully optimized mixture of 7B models has outperformed OpenAI’s GPT-3.5. Over a four-month period, their in-house LLMs have achieved a 20% day-30 engagement improvement, compared with the GPT-3.5 model. However, when the top models from the Chaiverse LLM competition were combined, day-30 engagement levels were elevated by 40% from the in-house models, marking a 68% total increase over GPT-3.5.
On the 21st of February, all existing and new users will have the option to upgrade their chat engine to Chaiverse v0.0.1, and the developers promise to deliver a much more compelling AI that users love.
How Do I Join Chaiverse As a Developer?
To join Chaiverse, one simply needs to join the Chaiverse Discord. A bot called “Chai AI” will greet you with a unique developer key. One must also download the Chaiverse pip package and use the developer key to submit models directly to users for real-world testing.
Chaiverse is currently said to be in beta mode. The Chai developers refer to each iteration of the Chaiverse product as a “season.” Externally, this is seen as a phase of the competition, where each season introduces a new target metric for developers to optimize towards. Since its launch, Chaiverse has received well over 10,000 unique model submissions, serving a range of model architectures such as LLaMa, Mistral, and Yi. Chaiverse has paid out to developers a total sum of $152,000 USD and promises to pay out $1 million USD before transitioning towards a direct monetization scheme.
