Why a specialized LLM should be special for your business

The term “vertical startup” inherently implies a narrow focus as these startups dig deep into a specific field or industry. They require specific AI models tailored to their narrow focus. Narrow AI, also known as specialized AI, is designed to handle tasks within a specific domain, as opposed to general-purpose AI, which aims to understand a wide range of human intellectual problems but may struggle with specialized knowledge. The demand for domain-specific AI models has increased with the rise of vertical startups in fields such as mathematics, finance, law, and healthcare.

BHSN is a vertical startup that specializes in the legal space. The startup was founded by practicing lawyers who combine their legal expertise with generative AI technology to create a unique legal AI service. Legal data is trained into their AI language model, and the AI helps review or revise contracts. The language model is allibee, a small-scale large language model (sLLM) specialized for legal purposes. allibee can analyze huge pages of complex contracts in seconds. It can identify problematic clauses and provide expert judgments, such as suggesting how a clause should be reworded. With this service, you no longer have to go through the traditional labor of sifting through page after page of contracts, comparing them to other contracts, or searching through numerous folders on your computer to find a mistake.

BHSN has redefined what they mean by “legal.” They unveiled their new vision at a press conference highlighting their technology. They emphasized that “legal” is integral to all business operations. For CEOs, it’s essential to base business planning, decision-making, and actions on legal documents such as contracts, legal advice, and legal reviews. In other words, they announced the principle that “Legal is business.” They are moving from a “narrow AI” to a broader AI, applying a legal-specific language model to most business decisions.

The beneficiaries will be companies that want to expand their business to Asia. English-speaking global companies looking to enter the Asian market can benefit from allibee’s assistance. One of the initial challenges for those looking to expand into another country is to verify the legal requirements for establishing a business in that country or reviewing contracts. However, legal documents in non-English languages can be challenging to interpret. It’s even harder to translate them. This is especially true in Asia, where many countries have their own languages, such as Korean or Japanese, making the language barrier a potential obstacle to market entry.

But what if a legal language model fluent in Asian languages could interpret local laws like an expert? BHSN’s “Business Intelligence,” set to be fully commercialized later this year, will be a B2B solution for companies looking to expand into Asia by providing expertise in local laws, government policies, case law, and news from various Asian countries. The Legal LLM is currently available in English, Korean, and Japanese, and there are plans to expand the range of services by developing AI language models specialized for Vietnamese and Chinese. This means that in the future, a Legal LLM will be able to answer your questions in English, even though it has mastered the local laws of Asia, eliminating the need for translation services.

For instance, if you ask allibee, “Can a foreigner invest in real estate and build a factory in Vietnam?” you will receive an immediate, clear answer based on local laws and procedures. allibee utilizes a powerful natural language processing engine called “Tokenizing*” to provide comprehensive responses. This shows the potential of Legal LLMs to offer guidance on business strategies and action plans based on legal considerations. This is the scalability of BHSN, from law to business.

These are some interesting points to consider. If the guidelines, codes of conduct, and work processes of every company were determined by the allibee Factory, could we slow down the countdown to the climate crisis? It’s not just about using less paper. We know that small, large language models (LLMs) have fewer parameters than general-purpose LLMs and that lightweight models can help reduce the time and cost of machine learning. Even if you don’t need a supercomputer, a small language model can still reduce your data center footprint. For these reasons, sLLMs are likely to have lower carbon emissions compared to LLMs. BHSN’s sLLM utilizes high-quality legal-specific data, fine-tuning, and a proprietary pre-training methodology* to enhance accuracy, competing with LLMs. A small, lightweight large language model may be small in size, but it can have a big impact.

Narrow is the widest, Small is the biggest. It may seem contradictory, but BHSN’s point is clear.