Weekly AI Recap: OpenAI unveils o1 ‘reasoning models,’ Meta to train AI using UK user data

The company's new models, dubbed o1, are designed to perform feats of abstract reasoning that push the company further along toward its goal of building artificial general intelligence, or AGI.

The company has released o1-preview and o1-mini, both of which are now available to ChatGPT Plus and Team subscribers. “We've developed a new series of AI models designed to spend more time thinking before they respond,” yesterday’s announcement reads. “They can reason through complex tasks and solve harder problems than previous models in science, coding and [mathematics].”

In its announcement, OpenAI included demo videos of o1 completing complex tasks. While solving one of those, a “logic puzzle” involving multiple variables that would require some time and serious mental effort from most humans, the model broke the problem down into multiple steps and provided a lengthy, detailed and ultimately correct answer.

OpenAI o1 solves a complex logic puzzle. pic.twitter.com/rpJbh8FkAg

— OpenAI (@OpenAI) September 12, 2024

Meta to start collecting data from UK adults to train AI

Social media giant Meta announced Friday that it will begin training its AI models using public content shared by adults in the UK across Facebook and Instagram.

“This means that our generative AI models will reflect British culture, history, and idiom, and that UK companies and institutions will be able to utilise the latest technology,” the company wrote in a blog post.

Private messages will not be collected to train Meta’s AI models, nor will public content from users under the age of 18, according to the post. “We’ll use public information – such as public posts and comments, or public photos and captions – from accounts of adult users on Instagram and Facebook to improve generative AI models for our AI at Meta features and experiences, including for people in the UK.”

Meta announced in June that it would pause the rollout of its AI models across the European Union after the Irish Data Privacy Commission (DPC) – its main privacy regulator in Europe – asked it to halt its practice of using public data from adult European users to train large language models (LLMs).

The company wrote in today’s blog post that since that pause it has “engaged positively with the [UK's] Information Commissioner’s Office (ICO)” – an independent data privacy watchdog – to develop a framework for legally training its AI models using adult user data in the region.

Stephen Almond, executive director of regulatory risk at the ICO, said in a statement Friday that the organization “will monitor the situation as Meta moves to inform UK users and commence processing in the coming weeks.”

EU privacy watchdog investigates Google’s AI training practices

In other privacy and AI news, the Irish DPC is investigating Google's use of consumer data to train its Pathways Language Model (PaLM) 2, Reuters reported Thursday.

The Irish DPC also recently filed court proceedings against the social media platform X after it was revealed that the company was training its AI chatbot Grok with user data obtained without consent. X shortly thereafter announced that it would stop training Grok with user data collected from the EU.

Taylor Swift underscores AI in Harris endorsement

Taylor Swift announced to her 283mn Instagram followers that she would vote for Kamala Harris in the upcoming presidential election, citing her “fears around AI” as a major contributing factor.

The singer-songwriter was the subject of AI-generated deepfakes which former President Donald Trump recently circulated online in an effort to win support from her vast and loyal fanbase.

Swift is just the latest A-lister to publicly voice their concerns about the dangers of AI being used to depict their likenesses or voice without consent. Such high-profile complaints, according to experts, will have a measurable impact on the broader social and political discourse swirling around the technology.

Adobe slated to release new text-to-video AI model

Tech giant Adobe announced in a blog post Wednesday that it will release its Firefly Video Model, capable of generating five-second video clips from text prompts, in beta later this year.

The brand says that the new model has been trained exclusively on licensed content, adding a layer of safety for marketers who are keen to experiment with AI-generated video but are also reluctant to use existing tools due to their associated copyright risks. The Firefly Video Model has been “designed to be commercially safe so marketers do not have to worry about infringing on other brands or intellectual property,” says Zeke Koch, vice-president of product management at Adobe Firefly.

The Drum speaks with Qatar Airways about its new AI marketing effort

Last week, Qatar Airways launched an ad campaign which allows viewers to place themselves center stage through the use of AI.

“AI Adventure,” as the campaign is titled, centers upon a three-minute film which tells the story of a man who travels the globe looking for a celebrity model with whom he briefly locked eyes and instantly fell in love during a chance encounter in London.

Viewers can submit photos of themselves via a microsite, after which point AI will impose their likenesses onto one of the characters’ faces, making it seem like they’re the star of the commercial.

Babar Rahman, the airline’s senior vice-president of global marketing, told The Drum in an interview earlier this week that the inspiration behind the campaign came from the desire to tell a particular kind of story, not from an effort to conspicuously wield AI. “The whole idea of the campaign,” he says, “was … how do we bring people into a world where they can bring travel and love together? AI was just an element which helped us deliver that narrative.”

For more on the latest happenings in AI and other cutting-edge technologies, sign up for The Emerging Tech Briefing newsletter.