Foundation models like ChatGPT and Stable Diffusion are large, pretrained AI models which, out of the box, come with amazing capabilities in terms of text and image generation. To be truly useful in a corporate context, however, they need to be further constrained and guided.
For text, the retrieval augmented generation (RAG) architecture has been the most common method of using these pre-trained models to ground them on the most relevant, up-to-date information and provide accurate answers in a corporate context. This type of setup feeds selected chunks of data to the foundation model via the prompt, without touching the model itself.
The last couple of months has seen more intensive efforts in the re-training or fine tuning of the foundation models itself. In this approach a dataset composed of company specific data and/or public data is assembled and used to adjust the model parameters slightly to better perform the specific task at hand.
In this webinar we are sharing our perspective on the benefits and hurdles of the fine tuning approach and provide you with practical guidelines on how to get started.
1. Foundation models: a snapshot of recent trends and news
2. Using foundation models: when to opt for model fine-tuning?
3. Fine-tuning foundation models: how to?
* You can access our previous webinar on Foundation models: ''Generative AI for corporate use: How to get started with LLM'' here.
Get access to the webinar by filling in the form below.