Prediction for 2024

Companies will own multiple fine-tuned LLMs/SLMs for specific tasks.

LLMs like ChatGPT are great for showing what these models are capable of doing in terms of breadth, but at the end of the day, you’re going to look for depth. Instead of relying on one-size-fits-all solutions, companies will look towards the integration of multiple, fine-tuned language models tailored for specific actions.

Enterprises are recognizing the importance of accuracy in their AI applications. General-purpose language models have been revolutionary, but the demand for specialized models is on the rise. From customer support interactions to complex data analysis, having dedicated language models for specific tasks enhances accuracy and efficiency. This is easy to understand since the weights that are being used for the LLM to have a big breadth of knowledge are repurposed for depth.

Personalization is no longer a luxury but a necessity. Specialized language models enable enterprises to deliver personalized experiences to their customers. We have to assume that everyone is utilizing the same models today, so offering ChatGPT in your product isn’t good enough. You need to add alpha to it. And that is done through fine-tuning utilizing your private data.

In an era of increasing cyber threats and stringent regulations, the deployment of fine-tuned language models allows enterprises to enhance their security measures and ensure compliance with industry standards.

As for the financial industry, this is 100% going to happen. Firms will fine-tune language models locally utilizing their proprietary datasets and providing access to their Snowflake/Elastic/ClickHouse/.. instances. This collection of models will effectively enhance the productivity of the firm by 2/3x, even displacing jobs.

We are preparing for this shift at OpenBB and I spent the last few days working on a proof-of-concept with José Donato that will allow users to bring their own copilots to the Terminal Pro. And even have these interact with each other.

For a video on how this works you can check: https://x.com/josedonato__/status/1741151037031845986?s=20