As the summer season heralded a series of legal disputes, OpenAI found itself amidst a tempest of lawsuits concerning privacy and copyright infringements. The core of these allegations is that OpenAI’s training of its expansive language models (LLMs) relied heavily on copyrighted content. But this narrative doesn’t only surround OpenAI; tech behemoths like Google, Meta, and Microsoft are similarly ensnared in related allegations.
Despite the legal maelstrom engulfing OpenAI’s LLMs, notably the GPT-4 and GPT-3.5 models, the undercurrents of concern appear to barely touch the legal tech entities that employ these models. A prevailing sentiment among these vendors is the strategic value of diversifying their AI models or championing an LLM-agnostic approach.
Legal Tech’s Calm Amidst the Storm
The sentiment in the legal tech corridor seems to be one of anticipated challenges with pioneering technology. Many believe that facing legal obstacles is par for the course, irrespective of their AI vendor affiliations.
Laura Safdie, representing Casetext, a company that leverages OpenAI’s LLMs, remarked, “We don’t perceive it as an immediate threat and we don’t hear the legal community raising those concerns because the nature of the legal questions that are being raised are usually about just the mechanics of training these foundational AI models.”
Yet, the horizon isn’t entirely clear for legal tech companies. There’s a latent risk lurking that could necessitate a pivot in the AI foundations of their products. Ken Crutchfield of Wolters Kluwer highlights the forward-thinking strategy of companies, suggesting they engineer their systems so that, “if that large language model … is not going to be an acceptable solution, for whatever reason, you can remove that and put in an alternative solution in that box.”
The Spectrum of Strategies in a Dynamic AI Landscape
The post-generative AI era presents a mosaic of strategies towards LLMs. Some firms broadcast a multifaceted approach, tailoring their solutions with a variety of models based on distinct requirements. Conversely, entities like Casetext have thrown their weight behind a single model. As Safdie elucidated, “The way we approach product development is we’re going to use the best model for the job. And so we’re constantly evaluating AI models, both public and nonpublic all the time.”
However, adaptability remains the industry’s watchword. Even those deeply committed to a single LLM are vigilantly surveying the horizon for emerging models. Praful Saklani from Pramata sheds light on the broader perspective: “The question really is, how can I use this in an ethical way? And how can I use this in a way that protects me from the invariable evolution? So we have a LLM independent mindset, and the lawsuits are part of that, but it’s also because who knows what the winner is?”
Lessons from the Tech Titans
Recent maneuvers in the tech juggernaut arena provide indicators of a paradigm shift towards a multifaceted AI approach. The recent collaboration between Microsoft and Meta, bringing Llama 2 to Azure, coupled with Amazon Web Services broadening its model suite, underscore this trend.
Saklani’s observations resonate with the larger tech zeitgeist, noting that many are “taking their cues from the people who understand AI best.” The emergent narrative suggests an industry transition away from a one-size-fits-all model, hinting at a variegated approach. As LLMs continue to evolve, the industry’s strategies may align in harmony with this evolution.
As Vishal Sunak from LinkSquares aptly summarizes the prevailing mood: “I think it’s still relatively unknown as to who has ‘the best approach to it’.” In this ever-transforming domain of generative AI, strategies are destined to be as dynamic as the technology they harness.
