0
0
New Legal Large Language Model Challenges Traditional Practices
OpenAI made headlines earlier this year with its claim that training leading AI models without copyrighted materials would be impossible. This assertion forms the basis of the company’s defense against accusations of mass data scraping for AI training. Critics argue for obtaining consent and paying licensing fees, while OpenAI defends its practices as fair transformative use.
Challenging this narrative is the introduction of KL3M (Kelvin Legal Large Language Model) by 273 Ventures. Founded by Daniel Martin Katz, a law professor at the Illinois Institute of Technology, and Michael Bommarito, the startup aims to revolutionize AI training without relying on copyrighted data, unlike competitors in the field.
KL3M: The First Certified Legal Large Language Model
Released in late February, KL3M recently received the distinction of being the first AI model to obtain a “Licensed Model (L) Certification” from Fairly Trained, an independent auditing organization. This certification validates the model’s compliant usage of data through contractual agreements or public domain access, setting a new precedent in AI training practices.
Training KL3M: A Revolutionary Approach
273 Ventures meticulously collected data from public domain sources such as U.S. government documents and legal filings to train KL3M. This approach, devoid of copyrighted materials, led to the creation of the Kelvin Legal DataPack, containing over 150 billion tokens. KL3M, trained on a curated subset of this dataset, offers two versions: kl3m-170m and kl3m-1.7b.
- KL3M-170m: 170 million parameters
- KL3M-1.7b: 1.7 billion parameters, optimized for high performance
With a focus on legal and regulatory applications, KL3M’s versatility transcends its intended industry, catering to diverse domains due to its broad knowledge base. Its performance benchmarks outshine competitors in terms of perplexity and toxic outputs, making it a viable solution for law firms and beyond.
KL3M’s Impact and Accessibility
Already in use by undisclosed law firms, KL3M’s adoption signifies a shift towards ethical AI training practices. While the model’s pricing remains undisclosed, interested parties can reach out to 273 Ventures for further details at [email protected].
Image/Photo credit: source url
