From Models to Infrastructure, Here's How to Navigate the AI Market
In the last decade, AI moved from science fiction to reality. The once-theoretical technology recently catapulted into the foreground of business and culture, largely thanks to revolutions in large language models, like ChatGPT. Companies are investing in AI at a remarkable rate, with private spend in the U.S. reaching nearly $50 billion in 2022, according to Stanford University’s 2023 AI annual report. Healthcare, fintech and data management are leading the way as they explore how this technology can help with everything from robotic process automation to virtual customer service agents.
AI INVESTMENTS IN 2022:
$6.1 billion
MEDICAL AND HEALTHCARE
$5.9 billion
DATA MANAGEMENT, PROCESSING, AND CLOUD
$5.5 billion
FINTECH
Nearly any business can profit from thoughtful AI integration so long as they are prepared to create the data engineering and governance foundations necessary to use the technology effectively. With so many types and models out there, it can be difficult to know where to start. Careful consideration of data infrastructure, company priorities and governance can help you determine which AI solutions are right for your organization at this time. Companies considering investing in AI need to make two choices: what model to use and how best to support it. Answering key questions at the outset can help narrow your scope and make those decisions easier.
POPULAR USES FOR AI IN 2022:
Choosing the Right Type of AI
Since OpenAI’s ChatGPT service released in November 2022, large language models (LLMs) have become the AI topic of conversation. Within six months, people visited ChatGPT 1.8 billion times. Together, they enter more than 10 million queries into the online portal every day. They use the tool to solve all sorts of personal, social and work-related problems, from writing sensitive emails to creating sales pitches and documenting code.
Large language models have the potential to revolutionize business, but that does not mean they are the most effective tools to meet your company’s needs. Adopting new technology is important, but it is even more important to use the right tool for the task at hand. Here’s a look at the strengths and weaknesses of some common AI solutions.
Large Language Models
EASY-TO-USE TOOLS FOR CONTENT GENERATION
ChatGPT and other LLMs can help marketers refine product descriptors and healthcare providers improve patient connections. These tools can boost engagement, increasing efficiencies and create a targeted experience.
• Most LLMs understand prompts in plain language, making them easy for everyone to use.
• LLMs provide writing assistance on a wide variety of subjects.
• Interpersonal communications: writing emails, prepping for a presentation or pulling key points from long notes • Creating routine documents: updating resumes, creating checklists and recording meeting minutes • Coding assistance: finding bugs in code, creating simple functions and generating documentation • Searching databases without the need for query languages (e.g. creating SQL queries)
• LLMs struggle when desired responses are too technical or specific. Companies should ensure they have the right people, computing power and resources to perform any additional model training that may be needed.
• Their ease of use can be a double-edged sword. Organizations run the risk of data leaks or other IP problems if people don’t use proper precautions. These leaks could be accidental, but it doesn’t make them any less dangerous.
MAIN TAKEAWAY
LLMs are quite powerful and can benefit almost anyone in a company. However, they are not the right solution for every problem. The purpose and use of an LLM should be carefully determined to maximize benefits and minimize risks.
Small Language Models
EXCELLENT SOLUTIONS FOR SPECIFIC PROBLEMS
Small language models are like large language models…but smaller. They are good at handling specific subjects or tasks, such as identifying parties in a legal contract, classifying documents or translating passages to another language or style of writing. For example, their narrowed scope can help retail companies reduce labor costs and human error while improving speed to customers with specific needs.
• SLMs are an easy and affordable option to train and maintain.
• They can be stored and run locally or in a company cloud.
• Companies can choose from many existing options – both open-source and proprietary – that they can plug in as is or after some simple model training.
• SLMs can be trained to generate content for very specific and technical purposes.
• SLMs do not have the ease of use or inherent flexibility of LLMs and often need more technical expertise to use.
• Data science, AI and data engineering experts are needed to help with model training and maintenance.
MAIN TAKEAWAY
Small language models can be excellent tools to create, edit or analyze writing and other content. However, they require specific training and expertise to maintain.
Other AI Tools
THE WORLD BEYOND LANGUAGE MODELS
AI can generate actionable insights from data beyond documents, from computer vision tools identifying supply chain problems to RPA hyperautomation of entire business streams. Industries like insurance can tap into these AI tools to generate accurate pricing and refine predictive models.
• AI can capture complex relationships in data that conventional statistical models may miss.
• AI tools help with a wide range of tasks outside of writing and content creation.
• Data classification • Computer vision • Forecasting • Robotic Process Automation • Much, much more
• These tools can be either trained from open-source models or used out-of-the-box from proprietary sources, depending on specific needs.
• Any AI solution requires expertise to conceptualize and implement, as well as data and computation power to train. This can require spending additional time and money on your solution.
• AI solutions can adapt to nearly any problem but may require specific skills to use effectively. As a result, your company may need to bring on a new role or skillset.
MAIN TAKEAWAY
AI is much broader than language models. Just because your organization does not need any help writing, managing or summarizing documents does not mean that AI cannot provide value.
Traditional Machine Learning and Statistical Solutions
SMALL-SCALE SOLUTIONS FOR DATA INFERENCE OR CLASSIFICATION
Sometimes simple solutions can solve problems just as effectively as more complex AI options. Random forest classifiers, clustering algorithms and even simple regressions can provide information to improve decision-making and increase productivity. Financial services companies can use statistical modeling to inform trading research.
• Statistical or simpler machine learning models can perform as well as AI models in prediction, forecasting and classification – as long as relations in data are not too complex.
• These models require less data to train than a full AI and can produce useful results more quickly.
• Statistical models and the machine learning solutions that use them are markedly less flexible than AI models and can only capture certain data relationships.
• Specific model choice is much more important for this reason. Data science and statistics experts can ensure that the methods employed are those best suited to your company’s needs.
MAIN TAKEAWAY
Not all problems require big, elaborate solutions. For a basic task, a simple tool may be just as effective as the cutting edge in technology.
Determining the Best Support Strategy
Once you decide what AI tools are appropriate for your organization, it is important to consider what infrastructure will best support them. First and foremost: will you use an internal or external support model? Companies can train and maintain AI in their own environment, or they can choose to use one of the many out-of-the-box toolsets available through cloud AI integration, such as AWS and Azure services.
Company-Maintained Models
INTERNAL OPTION PROVIDING GREATER CONTROL AND FLEXIBILITY
Maintaining AI does not have to mean creating something from scratch. Many open-source models can be adapted with proper data and training to meet your organization’s needs and protect against external threats.
• Internal LLMs provide more precise content control, which is especially important if many of your users are not tech-savvy or if your organization is creating materials with sensitive or proprietary information.
• Less likely to yield accidental data breaches • Easier and more secure access to organizational data
• Both LLM and other AI models can be trained to use company- and industry-specific language or match specific requirements.
• Internal maintenance requires AI, data engineering and data science expertise. New roles or additional hires might be needed.
• Model training on a cloud platform should be done in consultation with AI experts to avoid unnecessary costs.
• Having a locally maintained model (LLM or otherwise) does not prevent the need for security protocols!
• Training and protocols are essential to minimize the chance of malicious training injections and other risks. AI and security experts can help draft these best practices.
MAIN TAKEAWAY
Internally maintained models are ideal for organizations that have very specific or technical needs from generative AI and are willing to invest in technical expertise.
Custom solutions to achieve your goals
No matter what AI use cases you find interesting, what models you want to use or how you want to support it, Kforce Consulting Solutions can help set your organization on the path to useful AI integration.
