- AI Research Insights
- Posts
- AI News: Microsoft-Second largest human-translated parallel test set; CMU-propose to model music as a series of discrete notes; Meta-Holistic Trace Analysis (HTA)
AI News: Microsoft-Second largest human-translated parallel test set; CMU-propose to model music as a series of discrete notes; Meta-Holistic Trace Analysis (HTA)
Hi there, today we will be sharing some research updates from Deepmind, Stanford, Interesting Conclusion On Microsoft, Stable Diffusion, CMU, Rice University, Open AI, Cohere, Meta, and much other cool stuff. So, let's start...
Microsoft Translator: It announces its release of the second largest human-translated parallel test set, featuring 128 languages each having 2000 sentences translated with a document context without post-editing.
CMU: The future of generative art is Personalization. New paper enables us to quickly teach Stable Diffusion many new concepts at once, and combine them on the fly.
Rice University: Researchers propose to model music as a series of discrete notes upon which they can use autoregressive natural language processing techniques for successful generative modeling. While previous works used similar pipelines on data, such as sheet music and MIDI, the research group aim to extend such approaches to the under-studied medium of guitar tablature.
Stable Diffusion: AUTOMATIC1111's Stable Diffusion WebUI is one of the most powerful tools in the generative AI space. Stable Boy puts that power into GIMP 2.10 by calling into A1111 WebUI's API.
Open AI: New Paper from OpenAI on potential Threats from Large Language Models
Cohere: Does Cohere Deliver a 3X Better MIRACL? Recently, CohereAI boasted "3X better performance" in multilingual text understanding. A research group tested that claim by evaluating Cohere embeddings on MIRACL and found doesn't appear so... but Cohere's multilingual embeddings nevertheless yield impressive quality improvements.
Meta: Introducing Holistic Trace Analysis (HTA), an open-source project that makes it easy to understand and debug performance issues for large-scale distributed training jobs. Fully scalable to support state-of-the-art ML workloads.
Eleven Labs': Eleven Labs new speech generation model allows users to set gender, age, accent, pitch and speaking style, and synthesize an infinite variety of new voices. The voice synthesized by the system is different every time, even if the user uses the same parameter Create a sound, and you also get a sound that didn't exist before.
ChatGPT API is in the works and will be available soon. Sign up for waitlist here to be among the first to access it
— Mira Murati (@miramurati)
1:53 AM • Jan 17, 2023
How did we do?
Did you like today's newsletter? |