Writer’s new AI models are scary good at healthcare and finance tasks

San Francisco-based AI company Writer launched two specialized large language models (LLMs) tailored specifically for the healthcare and financial services industries on Wednesday, potentially reshaping how these highly regulated sectors adopt artificial intelligence.

The new models, Palmyra-Med-70b and Palmyra-Fin-70b, are now available as open-source offerings on major AI platforms including Nvidia, Baseten, and Hugging Face. Writer claims these specialized models significantly outperform larger, generalized AI models like GPT-4 in domain-specific tasks.

“In general, there are not enough domain-specific models because they’re simply [too] hard to build,” Waseem Alshikh, CTO and Co-Founder of Writer, told VentureBeat in an interview. “Those models require not just engineering, but a special type of data and a special type of expertise. You actually need experts to help you build those models.”

Specialized AI models outperform general-purpose counterparts in accuracy tests

The launch comes as industries grapple with how to leverage AI’s potential while navigating complex regulatory environments. Healthcare and finance, in particular, have been cautious adopters due to stringent compliance requirements and the high stakes involved in decision-making.

Writer’s healthcare model, Palmyra-Med-70b, has achieved an average accuracy of 85.9% across all medical benchmarks in zero-shot attempts, surpassing competitors like Med-PaLM-2, which achieved 84% accuracy after six attempts, according to the company.

“We have our healthcare industry specialist, Ziad, and he’s actually the doctor… he actually helped us build this model,” Alshikh said. “By literally running all the tests, we could act with the model. And then we use this interaction as a reward system to the model itself.”

Breaking new ground in finance: Palmyra-Fin-70b aces CFA Level III exam

In the financial sector, Writer’s Palmyra-Fin-70b model is the first AI model capable of passing the CFA Level III exam, often considered one of the most challenging tests in the investment management profession.

“If you look for financial use cases, I think that one of the biggest use case commercial debt and the model doing amazingly,” Alshikh explained. “The interaction with the financial models were different, because staff actually act as a financial expert by asking the question is how we train the models will not just provisional answer, we’ll ask a question back, and then create this customized plan for you.”

Open-source strategy: Paving the way for transparent and safe AI in regulated industries

In a notable departure from many AI companies’ closed-source approaches, Writer has made both models open-source. This decision aligns with a growing trend in the AI industry, following recent open-source releases from tech giants like Meta and Mistral.

“We believe open source [will] be the way for [customers] to scale,” Alshikh said. “Relying on black boxes in this huge part of your organization is going to be a very, very tough decision.”

The move towards domain-specific AI models could address longstanding concerns about AI safety and regulatory compliance in highly regulated industries. “The only way actually to address [AI safety] should be open source, because you cannot just work with a blog post to analyze the answer,” Alshikh said.

Writer plans to develop more domain-specific models in the future, with Alshikh hinting at upcoming models in the creative domain and the addition of multimodal capabilities to existing models.

The company’s focus on specialized AI models for regulated industries could give it an edge in the competitive enterprise AI market. However, Writer will face competition from both established tech giants and other AI startups targeting similar markets.

As the AI industry continues to evolve, Writer’s launch represents a significant development that could reshape how enterprises, particularly in regulated industries, approach AI adoption. By combining domain-specific expertise with an open-source strategy, Writer is betting on a future where specialized, transparent AI models become the norm in enterprise applications.