The growing discourse around AI technologies often zeroes in on their expansive potential and transformative impacts. However, Darren Oberst, CEO of Ai Bloks and the AI framework platform LLMWare, argues that smaller, localized versions of AI language models could be essential in tackling burgeoning concerns related to data privacy and the costs associated with these technologies. Speaking at the Forward Festival in Madison, Oberst shed light on how these compact models could be the unsung heroes of the AI world. Hosted by the MadAI group, a community of AI professionals in the Madison area, his insights offered a refreshing take on the AI landscape.
Oberst’s observations come at a time when the tech community is heavily fixated on monstrous models like ChatGPT. Despite their allure and evolving capabilities, industry heavyweights such as Sam Altman and Elon Musk have frequently sounded alarms about the rapid pace of AI development. Coupling this with the worries corporations have about their sensitive data being potentially compromised, Oberst paints a picture of an AI domain rife with looming challenges.
“All is not rosy in generative AI land; there are some real storm clouds on the horizon,” Oberst warned. He quickly pivoted to propose that smaller language models, or SLMs, could play a pivotal role in addressing these challenges. According to Oberst, SLMs share the same foundational mathematics as their larger counterparts, differing primarily in the number of parameters—or variables—that define their functioning.
To put this into perspective, while “mega models” operate with hundreds of billions of parameters, SLMs function with far fewer, typically ranging between 1 billion and 10 billion parameters. This significant reduction allows these models to run on medium to high-end laptops, entirely offline, ensuring a secure and private processing environment. This capability is particularly advantageous for handling sensitive data, such as health records, investigative material, and government information, minimizing the risk of exposure.
Oberst highlighted an important nuance: while these smaller models require cloud-based training, their operational phase is self-contained and does not necessitate further interaction with broader information ecosystems. Despite their smaller scale, Oberst asserted that SLMs could competently handle most tasks expected of large applications like ChatGPT, including delivering fact-based answers and performing basic analyses.
“My experience is that a small model can probably do 80% to 90% of what the ‘mega model’ can do, but you’re going to be able to do it at probably 1/100th the cost,” Oberst noted. This economic feasibility makes SLMs particularly suitable for niche applications. For instance, a small business or an academic team could take a model that operates at roughly 80% accuracy and tailor it to hit a 95% accuracy rate for specialized tasks.
Oberst passionately elaborated on this, saying, “The real promise of small models is not just, ‘Oh look, it can kind of do sort of what a big model can do.’ The idea is that because it is so much smaller and lower-cost to adapt and deploy privately, you can start fine-tuning these models. Instead of thinking, ‘I have one big model,’ I could have 10 smaller models, each tailored to perform a specific task or purpose.”
Oberst’s insights offer a compelling argument for rethinking the overarching emphasis on gigantic AI frameworks, encouraging a shift towards more versatile, cost-effective, and privacy-conscious SLMs.