At this year's ILTACON, amidst the usual networking and product pitches, a new trend was palpable among legal tech enthusiasts. Despite the ongoing struggles of AI giants like OpenAI and Anthropic to secure investor funding amid disappointing returns, whispers of a shift towards smaller, more efficient AI models were evident. These smaller models are not only proving to be cost-effective but are also offering comparable quality to their gargantuan counterparts, with the added benefit of potentially running on local systems— a significant advantage for data-sensitive sectors like legal.

This week, Meta stepped into the spotlight by unveiling its new small reasoning model, designed specifically for mathematical and coding applications and capable of operating locally. This move signals a broader industry pivot towards "tiny AI," a trend further underscored by Meta's recent mishap during a live demo of its general AI, which failed spectacularly, leaving many to question the viability of large-scale models.

Meanwhile, the Chinese AI model DeepSeek made headlines with its cost-effective performance, boasting operational costs and capabilities that challenge its American counterparts at a fraction of the price. DeepSeek's disclosed training cost of $294,000—a stark contrast to the massive expenditures of larger models—complemented by its competitive per-token pricing, positions it as a formidable player in the global AI market.

As for the concept of "agentic" AI, despite its buzzword status and the skepticism surrounding its practical applications, companies like Salesforce are heavily investing in this area. However, they report mixed success rates, highlighting the ongoing challenges in AI's ability to handle complex, multi-step tasks without human intervention.

Alibaba's AI research team also introduced Tongyi DeepResearch, a smaller model that matches the performance of OpenAI's larger systems but with significantly fewer parameters. This development not only showcases the potential of smaller models but also aligns with the industry's growing inclination towards leaner, more targeted AI solutions.

In the legal sector, where the need for vast computational resources is less critical than in other industries, smaller models offer an ideal balance between performance and practicality. They promise to enhance efficiency without the overhead of large-scale models, aligning closely with the industry's specific needs.

As the shift towards smaller AI models gains momentum, the implications for traditional heavyweights in the AI space could be profound. Just as the digital landscape has seen giants like Napster and MySpace replaced by more agile competitors, the AI industry may be on the brink of a similar transformation—a shift towards smaller, smarter solutions that better meet the specific needs of industries like legal. The future, it seems, may not be about who can build the biggest AI, but who can tailor it to be the most effective within real-world applications.