A couple of months ago, I did a presentation for my company where I demonstrated the state of Artificial Intelligence and how a business like ours could start taking baby steps into AI, ensuring that we are not left behind while also avoiding throwing the kitchen sink at it just yet. In this brief blog, I'll share my perspective on the current state of Artificial Intelligence as I observe, based on the information I gathered from sessions I attended at Big Data London 2024, various AI events across the United Kingdom, industry leaders I had conversations with, and obviously some research from the web, academic journals, and papers.
What is Artificial Intelligence (AI)?
Let’s start with the absolute basics, addressing a question that was more common a few years ago but is now often taken for granted, as we assume everyone already knows the answer.
The definition of AI from Oxford suggests, Artificial Intelligence is "the theory and development of computer systems able to perform tasks normally requiring human intelligence". In all essence this definition seems correct but are we at that stage of development yet? Answer is probably NO, but that doesn't mean we would never be, being an optimist, I feel we are closer now than ever have been. Hence, with the state of current development in mind, let's tweak the definition a bit. With the present proceedings, I would like to define Artificial Intelligence as " the theory and development of computer systems able to perform tasks normally human beings find repetitive and boring."
How does Generative AI help you now?
I view AI as a powerful problem-solving tool and have long believed that it offers numerous solutions and more efficient ways to tackle challenges involving language and textual data in general. So if you have a problem involving some textual data which you've been struggling with for long now, possibly GenAI has an answer to that now.
For instance, tasks like product categorisation using product descriptions, sentiment analysis for reviews in multiple languages, analysing survey data, or interpreting auction item descriptions—just to name a few—used to require toiling with code and tech. Now, these textual heavy tasks have become much simpler with the capabilities of Generative AI.
We now face a battle between fact and fiction, accuracy and hallucinations. Therefore, while Generative AI excels as a content generator, I believe it is still far from being capable of making decisions independently without a Human In the Loop.
Where do you stand in the Generative AI Pyramid?
This infographic caught my attention during one of the sessions at Big Data London 2024. The first question to consider is: where does your business sit on this pyramid? This assessment should factor in the data available, associated costs, technical expertise within the organisation, and potential use cases to determine the ideal starting point.
P.S. More often than not, the answer lies in the upper half of the pyramid. Contextual Prompt Engineering and Retrieval Augmented Generation (RAGs) tend to be the most effective for accomplishing tasks. Fine-tuning and training models, while powerful, often come with significant costs.
How to make the most of GenerativeAI for your Business?
Now that we understand AI can assist with repetitive tasks, excel in handling textual data, and we’ve assessed where our business stands in the pyramid, it’s time to outline key pointers to help us take those baby steps.
1) Prompt Engineering and RAG:
When getting started with Generative AI, designing task-specific prompts and learning to ground them in your own data with RAGs remain effective solutions to many problems. These techniques can often provide more value than you might initially expect and incurring significantly lower costs than fine-tuning or training models.
Here are some example posts from me that could be helpful for getting started on building AI applications with contextual prompt engineering and RAGs.
2) Human in the Loop:
As we’ve already established, AI at its current stage may not be a strong decision-maker capable of replacing human cognitive abilities. However, this doesn’t mean we should disregard its potential. The solution to addressing hallucinations and potential errors lies in incorporating a human-in-the-loop. This approach ensures relevant audits, sense checks, and the tracking of evaluation metrics to measure and compare models for factors such as relevance, groundedness, toxicity, and more.
3) Getting Data 'AI Ready':
If you're in the early stages of adopting Generative AI, it's crucial to adopt practices that ensure your data remains relevant for Large Language Models. Focus on maintaining meaningful naming conventions, simplifying datasets, and unifying data points for easy access. Additionally, ensure safety and security within the infrastructure, and be proactive about security and privacy terms before implementing any model or infrastructure. These key elements are essential for making data suitable for LLMs.
4) Leveraging Open Source with No/ Low Cost:
It's very easy to get billed from anywhere in this day and age, if you're a small-medium enterprise, it's of utmost importance to keep a tap on billings. Personally, I aim to solve problems at little to no cost by fully leveraging open-source resources. Whether it’s models like Meta’s LLaMA, Weaviate with its vector databases, or open-source frameworks like Langchain, there’s plenty of open-source help available. Businesses should avoid becoming overly reliant on platforms and continuously seek alternative solutions to keep costs low. As markets become more cost-sensitive, being able to do things at little to no cost gives you a significant advantage. Additionally, reiterating what I said earlier, It's not the time to throw kitchen sink just yet.
5) Product Mix:
In the industry, you’ve likely heard the mantra: "under-promise and over-deliver." Building on that, I’d categorise the core applications of AI into internal and external stakeholders (client-facing) use cases. By focusing on building and testing AI solutions internally first, you can ensure you’re fully prepared to confidently answer client question when it comes, "Could Generative AI possibly help with that?"
These are a few key insights that resonated with me during conversations with professionals working in Generative AI, insights from industry experts, and my own interpretation of how small-medium sized businesses can begin now to bridge the gap between knowledge and execution. All businesses should recognise that leveraging AI to simplify processes is gradually becoming the norm. In my opinion, it’s better to act on this knowledge sooner rather than later.
Feel free to comment or reach out to me—I’d love to hear your thoughts!
Signing Off,
Yash Sakhuja
Comments