Every decade or so, a significant technological platform shift occurs, with transitions from PC to web and then to mobile. In 2023, the landscape is undergoing another substantial platform shift with AI, and GPT-4's release in March marked the beginning of this transformation. I have fully immersed myself in this exciting development and couldn't be more thrilled about the opportunities it presents. The best thing I did was to build my own PC with an RTX 4090 graphics card to run large language models (LLMs). There are lots of good open-source tools available for downloading and running LLMs without having to write any code. After experiencing the convenience and efficiency of your own LLM, it's hard to imagine going back to the way things were before.

The captivating new releases of open-source large language models, the corporate drama surrounding OpenAI, exploring a new open-source language model daily, and building new AI projects have kept me engaged throughout the year. I am eager to share my thoughts on what 2024 may bring.

Pause and take a look at the graph above. There is a reason for not including any measurement such as months or years in it but simply showing the trajectory we are on. AI experts may differ in their timeline on when we will reach AGI but they all agree with the graph. The next 5 years is going to bring so many changes to what we believe as normal. 2024 is the start line.

What will we see in 2024?

2023 was the year when we built powerful AI Large Language models (LLMs). 2024 will be the year when we'll integrate these LLMs everywhere. Companies will start to become “AI First” and winners will arise from experimenting, adopting and building their own AI system rather than waiting on the sideline or using it from 3rd party providers.

Long Running Retrieval Augmented Generation (RAG) Workflows

Retrieval Augmented Generation (RAG) is a powerful technique that combines retrieval and generation capabilities to provide more accurate and contextually relevant responses.

Many people signed up for both free and pro accounts on OpenAI, Claude, Perplexity, and 10,000 other AI apps. Each app featured a chat box where users could type a prompt and quickly receive a response. This setup met some basic needs. However, the true value emerges when you can continuously process your emails, documents, and other information to obtain responses. Imagine running all your emails through an AI and having it create draft responses for the emails you are most likely to reply to.

Long-running RAG workflows can be particularly useful in scenarios where large amounts of data need to be processed, analyzed, and synthesized into coherent and meaningful insights.Moreover, long-running RAG workflows can be fine-tuned and customized to specific use cases, further enhancing their accuracy and relevance. This is simply not possible if you are using a LLM sitting in the cloud. You have to run it on your own locally.

Fine-Tuning Open-Source Models with Private Data

Open-source models are pre-trained machine learning models that are publicly available for use and customization. By fine-tuning these models with private data, organizations can tailor them to their specific needs and requirements, thereby improving their performance and accuracy.

Fine-tuning open-source models with private data can be particularly useful in scenarios where the data is highly specialized or unique to the organization. By leveraging the pre-existing knowledge and capabilities of the open-source model, organizations can quickly and efficiently build custom models that are optimized for their specific use cases.

Deploying Open-Source Models on Your Own Instance

Deploying open-source models on your own instance refers to the process of installing and running these LLMs on your own servers or infrastructure, rather than relying on third-party providers or cloud services.

By deploying open-source models on your own instance, organizations can maintain greater control over their models and data, as well as ensure greater security and privacy. This level of control can be particularly important in industries where data privacy and security are paramount, such as healthcare or finance.

Moreover, deploying open-source models on your own instance can provide greater flexibility and customization options, as organizations can tailor the models to their specific needs and requirements. This level of customization can be particularly valuable in scenarios where the models need to be integrated with existing business processes or workflows.

Ability to Build Apps That Work and Integrate it with Existing Business Processes

AI won’t replace humans, humans that use AI will replace humans. Every company has to become an “AI First” company to extract value and gain competitive advantage.

By building custom models that are optimized for their specific use cases, organizations can ensure that their models are accurate, relevant, and effective. This level of customization can be particularly important in scenarios where the data is highly specialized or unique to the organization.

Moreover, by integrating these models with existing business processes and workflows, organizations can ensure that they are seamlessly incorporated into their operations, thereby maximizing their impact and value. This level of integration can be particularly important in scenarios where the models need to be used in real-time or in conjunction with other systems or tools.

So, if you are preparing a list of To-Do items for 2024, then add “Get your own AI” to top of that list.

Happy New Year 2024..!