Generative AI

3. A well-known phenomenon in large language models, in which the system provides an answer that is factually incorrect, irrelevant or nonsensical, because of limitations in its training data and architecture.
6. A scenario where a model can perform tasks it hasn't been explicitly trained on. In generative models, zero-shot learning can involve generating data from categories not present in the training data.
7. Neural networks designed to work with sequences of data, where information from previous steps is carried forward. They are often used for text generation, speech recognition, and time series prediction.
8. A unique language modeling technique called "permutation language modeling" has been employed in the development of this LLM, jointly created by Carnegie Mellon University and Google

1. The act of dissecting text or other data forms into smaller components, known as tokens, is termed tokenization. In the context of language models, tokens encompass words, subwords, or characters.
2. An approach developed by Ian Goodfellow which employs a pair of neural networks, namely the generator and the discriminator, trained in a adversarial fashion.
4. A neural network architecture used for data compression and reconstruction. It consists of an encoder that maps input data to a lower-dimensional latent representation and a decoder that reconstructs the original data from the latent representation
5. A new generative AI assistant in SAP