Announcing Charlie 1: A new Large Language Model from GoCharlie

We’re excited to announce the completion of Charlie 1, our new large language model (LLM) that goes toe to toe with state-of-the-art offerings on the market. Earlier this year, we trained a domain-specific LLM for short-form marketing content, Charlie Beta, that was released to a subset of our customers. At the time, this model outperformed all existing models in 80% of blind user tests. We’ve brought learnings from that initial release into the development of Charlie 1.

Charlie 1 allows businesses to build exceptional AI applications across all business functions. Charlie 1 represents a significant leap forward in our AI capabilities and boasts both instruction and chat generation capabilities, the use of a system prompt, tuned for business-specific retrieval augmented generation (RAG) applications, reduced hallucinations, and is trained to be uncensored.

Instruct and Chat Capable

Charlie 1 is trained to handle both instructional and chat-style requests, providing unparalleled versatility and flexibility in how our customers can utilize the model. Charlie 1 can generate one-off content from a few-shot or zero-shot prompt or can be used for multi-turn chats to allow conversational solutions.

System Prompt

Charlie 1 includes system prompt capabilities that enable users to give custom instructions to how Charlie responds to suit the task at hand and the audience utilizing its capabilities. System prompts also give Charlie more personality, roles, and custom rules. This feature can make Charlie feel like an extension of your team to tailor responses as required by you or your organization.

Reduced Hallucinations

We enhanced Charlie 1’s reasoning capabilities, resulting in a 4X reduction in hallucinations relative to Charlie Beta and improved prompt-to-output alignment. This enables developers and businesses building on Charlie 1 or custom versions of Charlie to provide their stakeholders with a reliable experience. In our model development, we strive for significant reliability and precision.

RAG-Optimized

Charlie 1 is tuned for business-specific RAG business applications such as question-answering, key-value extraction, and summarization. For knowledge-intensive applications, Charlie 1 generates factual answers to queries on complex business documents, including contracts, financial reports, and legal documents.

Uncensored

Charlie is the first-ever uncensored commercial model, a feature that we believe allows the model to be better adapted to its end-use cases. That means no “As a language model, I cannot…” type outputs. We strive to develop truly useful AI, amplifying the unique version of you and your organization. We fundamentally believe that LLMs should be designed with maximum flexibility in how they can respond, where future safeguards can be added on top to filter the output to be more aligned with individual enterprises.

Come Join Us!

Our mission is to augment business functions by putting AI at the core of every business. Charlie 1 is a massive step towards the path of achieving this.

We plan to release more technical results of Charlie 1’s abilities soon. If you want to learn more about how Charlie 1 can benefit your organization, please reach out here.