Top AI News, July 2024: Llama 3.1, Mistral Large 2, Meta AI Studio, Search GPT

In this monthly roundup, we highlight the top AI news stories from July:

Canva x Leonardo AI

Canva has acquired the generative AI startup Leonardo.ai to enhance its AI capabilities and expand its Magic Studio suite. The deal, involving both cash and stock, integrates Leonardo.ai’s technology and team of 120 employees into Canva. Originally focused on video game assets, Leonardo.ai now offers tools for image creation across various industries and is known for its real-time photorealistic image generation through its Live Canvas feature. This acquisition aims to bolster Canva’s AI offerings and grow its user base, which already includes over 180 million monthly users. Leonardo.ai’s technology will be incorporated into Canva’s suite, potentially enriching its design and image generation tools.

Perplexity Shares Revenue With Publishers

Perplexity AI has introduced a revenue-sharing model for publishers following accusations of plagiarism from major media outlets like Forbes and Wired. The new “Publishers Program” involves sharing a percentage of advertising revenue with publishers whose content is cited in Perplexity’s AI-generated answers. Initial participants include Fortune, Time, and Der Spiegel. The program aims to enroll 30 publishers by the year’s end, offering API credits and analytics through ScalePost.ai.

The context behind: Condé Nast has sent a cease-and-desist letter to Perplexity, accusing it of plagiarism for using content from its publications like Vogue and The New Yorker in AI-generated responses without permission. The issue reflects the growing tension between AI companies and media organizations over content use and copyright issues, with some opting for partnerships rather than legal battles.

Suno and Udio Hire Elite Law Firm

AI music companies Suno and Udio have enlisted the elite law firm Latham & Watkins to defend against copyright infringement lawsuits filed by major labels Sony Music, Warner Music Group, and Universal Music Group. Latham & Watkins, which has previously defended major AI firms like Anthropic and OpenAI, will likely argue that using copyrighted material for training falls under the fair use doctrine.

The context behind: The labels allege that Suno and Udio unlawfully used their sound recordings to train AI models, potentially flooding the market with machine-generated music that could undermine original recordings.

X Trains Grok

X has faced backlash for automatically opting users into data sharing with Elon Musk’s AI startup xAI to train its AI assistant “Grok” without explicit consent. Users discovered the hidden default setting and can currently only opt out via the desktop version, with a mobile option in development. Grok 2 and Grok 3 are scheduled for release in August and December, respectively, with Musk claiming the latter will be the most powerful AI ever created.

Why it matters: The issue highlights the ongoing tensions around the use of user data for AI training without clear user consent.

Llama 3.1

Meta has released Llama 3.1, featuring the Llama 3.1 405B model, its largest and most advanced open-source AI model. With a context length of 128K and support for eight languages, Llama 3.1 offers capabilities in general knowledge, steerability, and multilingual translation.

Why it matters: This release is a significant step towards making high-quality AI accessible to developers globally, promoting innovation in areas like synthetic data generation and model distillation.

Mistral Large 2

Mistral has unveiled its latest model, Mistral Large 2, which marks a significant upgrade over its predecessor. With a vast 123 billion parameters, the model excels in code generation, mathematical reasoning, and multilingual support across 10 major languages and over 80 programming languages. Mistral Large 2 features advanced function calling capabilities and a 128k context window for handling long-context applications. It also demonstrates improved performance in reasoning and reduced “hallucination” tendencies. 

Between the lines: Mistral Large 2 has been released just a day after Meta’s Llama 3.1. Mistral claims that Large 2 sets a new standard for performance and cost efficiency in open AI models. With 123 billion parameters—less than a third of Llama 3.1’s 405 billion—Large 2 outperforms Llama 3.1 in code generation and math tasks, according to several benchmarks.

Instagram AI characters

Meta has launched AI Studio in Instagram, allowing users in the US to create AI versions of themselves to engage with followers. This tool enables creators and business owners to automate interactions such as chat threads and comment responses, customizable based on their Instagram content and preferences. Despite potential risks, Meta ensures these AI profiles are clearly labeled and provides guidelines for safe and controlled AI interaction.

The context behind: This move follows Meta’s September 2023 release of AI Characters, some of which were created in collaboration with well-known public figures.

SearchGPT

OpenAI announced a prototype of SearchGPT, designed to combine their AI models with real-time web information, providing fast and accurate answers with clear sources. Currently available to a limited group of users and publishers for feedback, SearchGPT aims to enhance search efficiency by delivering direct responses and follow-up answers in a conversational manner. The prototype emphasizes transparency, prominently citing and linking to sources, and aims to foster a symbiotic relationship with publishers by helping users discover high-quality content.

Apple Intelligence

Apple has released the first preview of its AI suite, Apple Intelligence, during the developer beta of iOS 18.1. Introduced at WWDC 2024, this suite enhances Siri, auto-generates emails and images, and sorts notifications. Initially available to developers for testing on iPhone, iPad, and Mac, users must register via a waitlist in Apple’s settings app to access more complex features. While the public release is slated for later this year, Apple Intelligence will be exclusive to iPhone 15 Pro models and newer, aiming to drive significant upgrades.

Behind the scenes: Apple is set to gain an observer role on OpenAI’s board, marking a significant step in their evolving partnership. Phil Schiller, Apple’s former marketing chief and current head of the App Store, will fill this position. As a board observer, Schiller can attend meetings but won’t have voting rights, allowing Apple to stay informed on OpenAI’s decision-making processes. This move follows Apple’s June announcement to integrate ChatGPT into iPhones, iPads, and Macs.

Spread the word

Posted

in

by