Image: © ink drop/Stock.adobe.com
The tech giant continues to focus on AI technology, while French start-up Mistral steps into the competitive market of large language models.
Meta has revealed a fresh batch of AI tools for its family of social media platforms, which include AI stickers for messages, an AI assistant and image-editing tools.
The range of upcoming features were revealed at Meta’s Connect event, where the tech giant also shared details of its mixed-reality headset – the Quest 3.
Meta said its AI stickers feature will be able turn text prompts into colourful, imaginative stickers, in a similar way text-to-image generators such as Dall-E operate. The feature utilises Llama 2 – Meta’s latest large language model – and its foundational model for image generation called Emu.
The company also revealed Meta AI, a chatbot-style assistant that will be able to answer user questions and help them find certain information. The tech giant also plans to launch 28 AIs in beta form, which different personalities.
“Some are played by cultural icons and influencers, including Snoop Dogg, Tom Brady, Kendall Jenner and Naomi Osaka,” Meta said. “We’re going to start rolling these out in beta in the US today. We’ll add new characters in the coming weeks played by Bear Grylls, Chloe Kim and Josh Richards, among others.”
Another upcoming feature is image editing with AI, which Meta plans to introduce to Instagram in the form of two tools. The first of these – Restyle – will lets users reimagine their images by typing in certain descriptions, such as ‘watercolour’ or other visual styles.
The second feature – Backdrop – will let users change the scene or background of their image through text prompts.
Meta has been pushing forward with its AI research. Earlier this month, the company released a tool that it claimed could detect biases – such as those relating to race and gender – within AI-powered computer vision systems.
Mistral launches its AI model
Meanwhile, a French start-up that had a highly successful launch earlier this year, has released its own large language model, jumping into competition with the likes of Llama and ChatGPT.
Mistral has launched its AI model called Mistral 7B, which the company claims is the most powerful language model for its size to date. The AI model contains 7.3bn parameters and has been released under an Apache 2.0 licence.
The start-up claims its model was able to outperform Llama 2’s 13bn parameter model on “all benchmarks” and also claims it can be easily fine-tuned to suit various tasks.
“We’re looking forward to engaging with the community on ways to make the models finally respect guardrails, allowing for deployment in environments requiring moderated outputs,” Mistral said in a blogpost.
The company was founded by former researchers from Google’s Deepmind and Meta and has ambitious plans to create open-source large language models, as a counter to others like OpenAI in the rapidly growing market.
One month after it was founded, Mistral managed to raise €105m in a seed round, showing the current hype surrounding the AI sector.
10 things you need to know direct to your inbox every weekday. Sign up for the Daily Brief, Silicon Republic’s digest of essential sci-tech news.
