Meta Releases New Research Models

Meta’s Fundamental AI Research team publicly released AI models, including image-to-text and text-to-music generation models, a multi-token prediction model, and a technique for detecting AI-generated speech, aiming to accelerate future research and responsible AI innovation

All this and more in today’s newsletter

Featured Byte

Meta’s Fundamental AI Research (FAIR) team recently publicly released advanced AI models to accelerate research and foster innovation in the field of artificial intelligence.

1. Chameleon Models: Meta introduced the Chameleon family of mixed-modal models that can process and generate both images and text simultaneously. Unlike most large language models, which typically produce unimodal results (e.g., turning text into images), Chameleon can handle any combination of text and images as input and output. Imagine generating creative captions for images or creating entirely new scenes using a mix of text prompts and images.

2. Multi-Token Prediction: To build better and faster large language models (LLMs), Meta proposed a new approach called multi-token prediction. Instead of predicting one word at a time, these models predict multiple future words simultaneously. The pretrained models for code completion using this approach are now available under a non-commercial, research-only license.

3. JASCO for Music Generation: Meta’s generative AI capabilities extend to music. The JASCO model offers more control over AI music generation, allowing users to turn text prompts into music clips.

News Bytes

• YouTube has introduced an updated “Erase Song” tool for creators, allowing them to easily remove copyrighted music from their videos without affecting other audio elements like dialogues or sound effects. The new tool employs an AI-powered algorithm to specifically detect and remove the copyrighted song while preserving the rest of the audio.

• OpenAI recently faced two significant security issues. Firstly, the Mac app for ChatGPT stored user conversations locally in plain text, making them vulnerable to other apps or malware. After this was exposed, OpenAI added encryption to the stored chats. Secondly, in 2023, a hacker accessed internal messaging systems, raising concerns about the company’s security practices.

• Chinese tech companies, both industry giants and ambitious startups, gathered at the World AI Conference in Shanghai this week. Despite facing U.S. sanctions, they showcased their latest innovations and expressed strong support for the country’s artificial intelligence sector. Notably, SenseTime unveiled its SenseNova 5.5, a rival to OpenAI’s GPT-4o in mathematical reasoning.

• Samsung Electronics is expected to achieve a remarkable 13-fold increase in second-quarter profit compared to the previous year. This surge is driven by growing demand for artificial intelligence (AI) technology, which has led to a rebound in memory chip prices. The tech giant’s semiconductor division benefited from rising memory chip prices, especially for high-end DRAM chips used in AI chipsets and data center servers.

AI Image Generators Products for Productivity

• MidJourney V6 : Best in class image generator based on text description

• Stable Diffusion : Latent text-to-image diffusion model capable of generating photo-realistic images given any text input

• Adobe Firefly 3 : Powerful AI tools for easy image editing, generation and enhancement

• DALL·E 3 : It allows you to easily translate your ideas into 
exceptionally accurate images.

• Ideogram AI : Image generator capable of creating impressive illustrations and integrating text.

Feedback Bytes

Thanks for reading and If you’d like to sign up for this newsletter or share it with a friend or colleague, you can find us right here. 📬