Generative Art Controversies, Speech Synthesis Fundings, and RLHF vs SFT vs IFT

Also, Introducing Hugging Face Competitions

Fabio Chiusano
January 30, 2023

Here are your weekly articles, guides, and news about NLP and AI chosen for you by NLPlanet!

Shutterstock has rolled out a generative AI toolkit to create images based on text prompts, while Getty Images is currently embroiled in a lawsuit against Stability AI.
ElevenLabs launches Beta platform for creators/publishers to narrate long-form content. Powered by deep learning model for speech synthesis, it adjusts delivery based on context.
What are RLHF, SFT, and IFT, and how are they used in ChatGPT? Usually, the language-modeling objective of the base model is not sufficient for a model to learn to follow a user’s direction in a helpful way.
Announcing Hugging Face Competitions, that let you create public/private competitions with full control of datasets, metrics and evaluation.
Using recursive GPT prompts to generate SQL queries from natural language descriptions.

New lessons Question Answering and Text Summarization are out. New lessons will be published on February 6th.

Introducing MusicLM: Generating Music From Text. MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task and generates music at 24 kHz. Experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description.
MAV3D (Make-A-Video3D), a method for generating 3D dynamic scenes from text descriptions.
Can GPT-3 be used as a good data annotator for NLP tasks? The authors investigate and analyze its potential as a general-purpose data annotator in NLP.
Why do tree-based models still outperform deep learning on tabular data? Tree-based models remain state-of-the-art on medium-sized data (∼10K samples) even without accounting for their superior speed.
Survey of Hallucination in Natural Language Generation. Large language models-based generation is prone to hallucinate unintended text, which degrades system performance and fails to meet user expectations in many real-world scenarios. The authors provide a broad overview of the research progress and challenges in the hallucination problem in NLG.
Structured Prompting: Scaling In-Context Learning to 1,000 Examples. The authors introduce structured prompting that breaks the length limit and scales in-context learning to thousands of examples. Demonstration examples are separately encoded with well-designed position embeddings, and then they are jointly attended by the test example using a rescaled attention mechanism.

GPT Index, a way to make it easier to use large external knowledge bases with LLMs. It removes concerns over prompt size limitations and abstracts common usage patterns to reduce boilerplate code in your LLM app.

Thank you for reading! If you want to learn more about NLP, remember to follow NLPlanet. You can find us on LinkedIn, Twitter, Medium, and our Discord server!