Writer, an expert in LLM who takes the opposite view of OpenAI

Writer emerges as a bold player in the field of large language models (LLM). While OpenAI establishes itself with its generative solutions, Writer bets on a radically different and innovative approach. By focusing on vertical models tailored to specific sectors, this company disrupts traditional paradigms.
Reduced training costs and superior response quality define its strategy. Data optimization through advanced techniques allows Writer to directly compete with giants like GPT-4. At the crossroads of efficiency and ethics, Writer announces a future where model personalization will no longer just be an option, but a necessity.

Based in San Francisco, the company Writer stands out as an innovative player in the field of large language models (LLM). Since its creation in 2020, it has successfully raised 326 million dollars, attracting attention for its unique approach focused on verticalization of language models. Unlike OpenAI, which focuses on generalized models like ChatGPT, Writer develops specialized LLMs tailored to various sectors, such as retail, finance, healthcare, and customer support.

An innovative training method

Writer has designed its own LLM, named Palmyra, which includes 20 billion parameters. This model was trained on 800 billion tokens, mostly composed of synthetic data. This approach allows for circumventing the high costs associated with manual labeling of training information, estimated at 100 million dollars for GPT-4. Writer’s investment in its model amounts to only 700,000 dollars, a significant difference that reflects its effective strategy.

Kev Chung, Chief Strategy Officer of Writer, emphasizes that “creating an LLM that costs more than it brings in makes no sense.” This statement highlights the importance of efficiency and precision in AI model development. Writer manages to deliver more effective results through its verticalization strategy while minimizing costs.

Data supporting performance

To develop its LLMs, Writer also relies on licensed training data, enabling the justification of the relevance and robustness of the results obtained. This strategic choice aims to limit the risk of hallucinations – a recurring problem with some OpenAI models.

Writer also adopts self-evolving models, contrary to traditional static systems. These models integrate new information over time, improving their accuracy and relevance. They are based on three key elements: integrated memory, learning based on uncertainty identification, and an autonomous knowledge updating process.

Augmented retrieval generation

An innovative graph-oriented augmented retrieval generation (RAG) mechanism is also under development. Through this process, Writer manages to enhance the accuracy of responses generated from its clients’ documentary bases. By exploiting semantic relationships, Writer performs robust and fast analyses, transforming data into actionable information.

A vision of workflow automation

Writer envisions a future where LLMs will become a commodity. The company plans to offer a customizable platform, allowing clients to adapt language models with their own data. Kev Chung emphasizes the importance of integrating these models into the mainstream applications used by its clients, such as Salesforce and other large-scale software.

The advancements achieved by Writer experiment with the agentic revolution, where models connect to various third-party applications to automate workflows. For example, in marketing, it may be possible to create a product and manage all associated content through seamless integration with tools like Salesforce.

A prestigious clientele

Writer has already succeeded in attracting a large number of clients, including renowned companies such as Accenture, Goldman Sachs, and Jaguar Land Rover. The company is also counting on the collaboration of the French company L’Oréal, which is deploying Writer’s solutions across its different regions. With 500 employees, Writer has internationalized, with offices in New York, London, and Singapore.

With growing ambitions, Writer focuses on research and development, particularly in enhancing self-evolving models to strengthen their efficiency. Writer’s wish is to harmoniously integrate with the applications that its clients use daily, thereby maximizing the benefits of new technologies.

Questions and answers about Writer, an expert in LLM who takes a different approach from OpenAI

What is Writer and how does it differ from OpenAI?
Writer is a company specialized in developing vertical language models, offering solutions tailored to specific fields such as retail, finance, and healthcare. Unlike OpenAI, which offers generalized models, Writer focuses on efficiency and precision by creating specialized LLMs.
What are the advantages of using vertical LLMs developed by Writer?
Writer’s vertical LLMs are designed to provide more precise and effective results based on the specific needs of industries. This helps reduce errors and improve the relevance of the responses provided by adjusting the model to application contexts.
How does Writer address cost issues in the development of its LLMs?
Writer emphasizes cost optimization by using synthetic data to train its models, limiting learning expenses to around 700,000 dollars, while avoiding the high costs associated with manual data labeling.
What technologies does Writer use to limit the risks of hallucinations in its LLMs?
Writer uses licensed training data to ensure that the results are relevant and controlled, thereby reducing the risk of hallucinations, which are incorrect or inaccurate responses generated by the model.
How do the self-evolving models developed by Writer work?
Writer’s self-evolving models adapt over time by integrating new information. They rely on integrated memory, learning based on uncertainty identification, and an autonomous knowledge updating process, which improves their accuracy.
What does Writer’s graph-oriented augmented retrieval generation (RAG) strategy consist of?
This strategy improves the accuracy of answers generated from clients’ document bases through the analysis of semantic relationships. This enables more robust and faster analysis for even more relevant results.
Who are Writer’s main clients and in which sectors do they operate?
Writer’s clients include large companies such as Accenture, Goldman Sachs, Nvidia, and L’Oréal, operating in various sectors like finance, retail, and healthcare.
What research and development objectives does Writer aim for in the future?
Writer focuses on integrating its platform with applications used by its clients, such as Salesforce and Adobe, and envisions advancements in the development of self-evolving models to increase efficiency and accuracy.
How does Writer envision the future of LLMs?
Writer anticipates that LLMs will become a commodity, driving it to develop a platform allowing users to customize language models with data specific to their needs.