SLMs: Big savings in tiny packages

• Author:
• Author name

  Quantumrun Foresight
• February 20, 2025

Insight summary

Small language models (SLMs) rely on fewer resources while supporting tasks like personal tutoring and specialized analytics, drawing attention to offline AI solutions. They may speed up product cycles and reduce costs in many fields, though privacy and job security remain concerns. Everyone could benefit from more efficient AI, but they also need to stay alert to the risks of automated decisions.

SLMs context

SLMs are artificial intelligence (AI) systems designed to handle natural language tasks with a more compact size than large language models (LLMs) like OpenAI’s ChatGPT. SLMs generally operate with a few billion parameters or fewer, making them easier to deploy in offline settings or on devices that do not have significant memory capacity. In 2023, AI company Databricks introduced Dolly with 6 billion parameters and trained it on targeted data to deliver more refined answers. This approach shows how focusing on narrower bodies of text can help address specific industry needs, such as those found in insurance or law.

SLMs use compression strategies that remove unnecessary parameters and reduce resource demands. Techniques like pruning and knowledge distillation trim the original LLM while preserving key functions and performance. Additionally, many SLMs rely on subsets of carefully selected data, allowing them to produce answers that are aligned with domain-specific applications. In 2024, Microsoft unveiled its Phi-3 family, which includes a 3.8-billion-parameter model known as Phi-3-mini, enabling more organizations to deploy AI without making large investments in infrastructure.

Multiple companies have been releasing their versions of SLMs, including Meta's Llama and Mistral AI's Ministral series. These compact models can be used on private servers to maintain data security, reducing reliance on extensive cloud-based systems. Notably, Databricks revealed that Dolly required just three hours of training on a single machine for USD $30, indicating that smaller size does not always lead to lower-quality outcomes. As research continues, SLMs are expected to play a central role in future AI applications across healthcare, finance, and manufacturing.

Disruptive impact

Individuals may use SMLs as personal tutors for language learning or creative writing, offering direct assistance without depending on large-scale services. People can also rely on them for personal budgeting tools, although they may receive inaccurate financial advice. However, privacy concerns may grow if users share sensitive data because of potential unauthorized access. Finally, SLM-based chatbots can ease loneliness by offering basic social interaction, but they might supply incomplete solutions or overlook deeper emotional needs.

Companies may integrate SLMs into data analytics workflows, allowing product development teams to spot unusual consumer trends more rapidly. They can also explore new pricing structures by offering subscriptions to secure, offline SLM platforms tailored to specialized fields. In addition, SLM-driven translations may help businesses communicate across languages, but they could produce errors when handling nuanced legal or technical documents. Leaders might also employ SLM-based workflows to cut operational costs, but they risk losing unique human insights if they rely too heavily on automated responses.

Meanwhile, governments may direct funding toward dedicated research centers for SLM development, allowing local institutions to focus on supporting social programs or infrastructure projects. Regulators could also propose clearer guidelines on how data is stored and processed when models handle tasks for public services. International negotiations may shift as policymakers debate fair rules regarding cross-border AI collaboration. Officials might also examine how to use SLM-based systems for citizen engagement, balancing rapid access to information with concerns about bias and misinformation.

Implications of SLMs

Wider implications of SLMs may include: 

• Human resource departments deploying SLM-based interview analysis, leading to less subjective evaluations and narrower candidate pools.
• Industrial design companies adopting SLM-driven mockup generation, leading to shorter product cycles and frequent design refreshes.
• Local newspapers relying on SLM writing assistants, leading to fewer entry-level editorial roles in the market.
• More businesses shifting to SLMs due to its cost-effectiveness and lower carbon footprint.
• Financial consultancies using SLM for portfolio optimization, leading to cost savings but diminished roles for junior analysts.
• Automotive firms integrating SLM into maintenance diagnostics, leading to new training standards for vehicle technicians.
• Universities employing SLM tools for grading essays, leading to varied concerns about academic integrity oversight.
• Small accounting firms using SLM software for tax preparation, leading to less reliance on large-scale service providers.

Questions to consider

• Which new skills could you develop if SLM-powered systems replace some of the routine tasks in your field?
• How could your local government balance SLM-based public service improvements with privacy and fairness concerns?