Startup Launches the AI Model Which ‘Never Hallucinates’
Pursuing artificial general intelligence (AGI) is a continuing endeavor in the field of artificial intelligence (AI). But although OpenAI is pushing for AGI, Writer’s co-founders May Habib and Waseem Alshikh have different viewpoint. Habib claims, “If you can unplug it, it’s not AGI.” About ten years ago, Waseem Alshikh experimented with then-emerging machine learning approaches to summarize massive chunks of information. This is when Writer’s CEO, May Habib, stumbled onto his work on GitHub. Today, with the development of deep learning models and transformers, Alshikh’s work has been amplified and forms the foundation of their generative AI firm, Writer.
Palmyra: The Large Language Model
Writer employs its large language model called Palmyra, named after an ancient Syrian city. Palmyra helps businesses and workers to produce and modify material, including emails, papers, advertising, and summaries that follow a company’s editorial requirements. According to Habib, Writer’s most recent linguistic AI model will “never create anything factually incorrect” because of its architecture, which places accuracy over creativity. This contrasts with most generative AI models that “hallucinate” and spew false information.
Writer models are created utilizing publicly available material and company-specific data, including PDFs, editorial-style manuals, and brand phrases. They are trained with 30 billion parameters. The model is customized for each business, and the data is kept in that business’ cloud. Habib says, “The data is used like an index, and we have this path to explainability that shows where specific facts came from.”
Competition in the Market
Other generative AI businesses increasingly focusing on the enterprise market directly compete with Writer. Venture capitalists, including Insight Partners, Upfront Ventures, and others, have invested $26 million. In contrast, Jasper’s unicorn copywriting AI tool has $125 million in venture money. According to the company’s co-founders, Writer’s key distinction is deploying its vast language models trained on company-specific data (unlike Jasper, based on OpenAI’s GPT-3.5).
As per Alshikh, “Most of the companies in the market are only playing in the application layer, and they don’t even do it well. They use someone else’s technology and put a UI on top of it,” adding “Jasper is an example. As an AI company, how will you control the output quality when you don’t control your AI itself?”
Encoder-Decoder Models: The Deciding Factor
The Writer employs machine learning, NLP, and transformers to comprehend and produce more of the material. The Writer combines encoders, a crucial component of a text-understanding transformer, and decoders, responsible for text prediction and generation.
Reddy, a Conviction investor who formerly worked with the generative AI search engine company Neeva, concurs that encoder-decoder models aren’t perfect but are superior to ChatGPT, at least in terms of accuracy.
“Encoder-decoder models do tend to have much lower hallucination rates than decoder-only models, which is the OpenAI GPT-3 structure,” says Reddy.
Accuracy and Inclusiveness: Writer’s Hallmark
In contrast to previous models, the Writer’s models enable communication between encoders and decoders. Once the encoder has comprehended the query, it extracts the data from the database that the client organization has provided. Furthermore, it instructs the decoder on how to build a response based on it. As a consequence, the Writer can use inclusive language and be truthful.
Habib asserts, “This is how we’re able to be accurate and write in an inclusive language.”
According to Habib, the firm wants to make it possible for businesses to create content consistent with their brand’s voice and tone while adhering to legal and regulatory requirements. Writers may produce more accurate material that sticks to corporate standards by training models on company-specific data.
In conclusion, the AI community has taken notice of Writer’s strategy of deploying its sizable language models and encoder-decoder architecture. With its emphasis on accuracy and inclusivity, Writer has the potential to change the business world and dominate the field of generative AI and AGI.