Tailoring the Platform Model to Specific Tasks

[numberlist]

How does the platform model work? In the language domain, the Grounded Model is trained by feeding it a large number of sentences (terabytes of data). The model will try to predict the last word of the sentence based on the words it has seen before.


This generative ability of the model, to predict and generate the next word based on the previous words it has seen, is why the Foundational Model is actually part of a field of AI called Generative AI . While these models are trained to perform, at their core, a generative task, predicting the next word in a sentence, we can actually adapt them to perform traditional NLP tasks, such as classification or named entity recognition.


This process is called tuning , where you can germany whatsapp number data adjust your Foundational Model by feeding in a small amount of labeled data. If you don't have data or only have very few data points, you can still use these Foundational Models , and they work really well in low-label data domains.


In a process called prompting or prompt engineering , you can apply these models to some of the exact same tasks. Advantages of Platform Model Performance: These models have seen a lot of data, so when applied to small tasks, they can outperform a model trained on only a few data points.


Productivity: Through prompting or tuning, you need much less labeled data to get a task-specific model than if you had to start from scratch. Disadvantages of the Platform Model Computational Cost: These models are expensive to train and run inference, making it difficult for small businesses to train a Platform Model on their own .