Facebook owner Meta said on Friday it was releasing a batch of new AI models from its research division, including a “Self-Taught Evaluator” that may offer a path toward less human involvement in the AI development process.

The release follows Meta’s introduction of the tool in an August paper, which detailed how it relies upon the same “chain of thought” technique used by OpenAI’s recently released o1 models to get it to make reliable judgments about models’ responses.

That technique involves breaking down complex problems into smaller logical steps and appears to improve the accuracy of responses on challenging problems in subjects like science, coding and math.

Meta’s researchers used entirely AI-generated data to train the evaluator model, eliminating human input at that stage as well.

The ability to use AI to evaluate AI reliably offers a glimpse at a possible pathway toward building autonomous AI agents that can learn from their own mistakes, two of the Meta researchers behind the project told Reuters.

Many in the AI field envision such agents as digital assistants intelligent enough to carry out a vast array of tasks without human intervention.

Self-improving models could cut out the need for an often expensive and inefficient process used today called Reinforcement Learning from Human Feedback, which requires input from human annotators who must have specialized expertise to label data accurately and verify that answers to complex math and writing queries are correct.

“We hope, as AI becomes more and more super-human, that it will get better and better at checking its work, so that it will actually be better than the average human,” said Jason Weston, one of the researchers.

“The idea of being self-taught and able to self-evaluate is basically crucial to the idea of getting to this sort of super-human level of AI,” he said.

Other companies including Google and Anthropic have also published research on the concept of RLAIF, or Reinforcement Learning from AI Feedback. Unlike Meta, however, those companies tend not to release their models for public use.

Other AI tools released by Meta on Friday included an update to the company’s image-identification Segment Anything model, a tool that speeds up LLM response generation times and datasets that can be used to aid the discovery of new inorganic materials.

© Thomson Reuters 2024

Meta, the parent company of Facebook, unveiled a set of innovative AI models developed by its research division on Friday, reported Reuters.
Among the standout tools is the “Self-Taught Evaluator,” which may reduce the need for human involvement in the AI development process. This development is an important step towards creating AI systems capable of learning from their own mistakes, potentially paving the way for more autonomous and intelligent digital agents.
In addition to the Self-Taught Evaluator, Meta also released updates to its image-identification Segment Anything model, a tool for accelerating response generation times in large language models (LLMs), and datasets designed to support the discovery of new inorganic materials.
First introduced in an August research paper, the Self-Taught Evaluator uses the same “chain of thought” technique employed by OpenAI‘s latest models. This approach involves breaking complex tasks into smaller steps to increase accuracy in fields like science, coding, and mathematics.
Crucially, Meta’s researchers trained the evaluator entirely on AI-generated data, eliminating the need for human input during the training phase.
According to Meta researchers, the ability of AI to evaluate other AI models accurately opens new possibilities for autonomous AI systems that can self-improve. This could lead to the development of digital assistants capable of performing a wide range of tasks without human intervention.
Self-improving AI models may also reduce reliance on the costly and time-consuming process of Reinforcement Learning from Human Feedback (RLHF), which involves specialised human annotators verifying data and checking AI-generated answers for accuracy.
Jason Weston, one of Meta’s researchers, expressed hope that as AI becomes more advanced, it will become increasingly capable of verifying its own work, surpassing human accuracy.
Other companies, such as Google and Anthropic, have also been exploring the concept of Reinforcement Learning from AI Feedback (RLAIF).
However, unlike Meta, these companies have been more cautious in releasing their models to the public.

New Delhi: Social media giant Meta has laid off employees across multiple teams, including at WhatsApp and Instagram, as part of an apparent restructuring exercise, reports said on Thursday. 

Meta did not immediately comment on how many employees were impacted in the latest round of job cuts.

In a statement to TechCrunch, the company confirmed multiple teams were hit by layoffs.

“A few teams at Meta are making changes to ensure resources are aligned with their long-term strategic goals and location strategy,” a Meta spokesperson was quoted as saying.

“This includes moving some teams to different locations, and moving some employees to different roles. In situations like this when a role is eliminated, we work hard to find other opportunities for impacted employees,” the spokesperson added.

Employees working on Reality Labs, Instagram and WhatsApp were affected by the layoffs, according to reports.

Several Meta employees took to social media platforms to announce they have been laid off.

Jane Manchun Wong, who was part of the Threads team, posted: “I’m still trying to process this but I’m informed that my role at Meta has been impacted. Thank you to everyone, especially my Threads and Instagram teammates, for my wild journey at Meta”.

“If anyone is interested in working together esp on software/security engineering, please reach out via my email, LinkedIn, etc noted on my personal website,” she wrote on Threads.

Earlier this year, Meta laid off employees at its Reality Labs division.

The social media company run by Mark Zuckerberg first laid off 11,000 employees in 2022. In 2023, Meta laid off another 10,000 employees and withdrew 5,000 open roles it had yet to fill, as part of CEO Zuckerberg’s “year of efficiency”.

Meta announced the expansion of Meta AI, the company’s in-house artificial intelligence (AI) chatbot, to six new countries on Wednesday. The UK has been included in the list of countries which are getting access to the AI chatbot now. The social media giant also highlighted that the AI platform will be expanded to 15 more countries soon, taking the number of total supported countries to 43. Apart from introducing Meta AI within its social media apps, the company also added voice-based AI capabilities to the Ray-Ban Meta smart glasses in the UK and Australia.

Adding an update to its July post, where Meta AI was first expanded to new regions including India, the social media giant stated that the AI chatbot has been rolled out to six new countries — Brazil, Bolivia, Guatemala, Paraguay, Philippines, and the UK. Additionally, support for the Tagalog language which is spoken in the Philippines was also added. To use the AI assistant, users in these countries can simply update Facebook, Instagram, and WhatsApp and they should see the new Meta AI icon.

Additionally, the company also added that the chatbot will be expanded to 15 more countries soon including Algeria, Egypt, Indonesia, Iraq, Jordan, Libya, Malaysia, Morocco, Saudi Arabia, Sudan, Thailand, Tunisia, United Arab Emirates, Vietnam, and Yemen. It is believed that Meta AI will be rolled out to these locales before the end of the ongoing year. The AI will also be updated to support Arabic, Indonesian, Thai, and Vietnamese languages.

The UK and Australia are also getting Meta AI capabilities on the Ray-Ban Meta smart glasses with the current rollout. However, the company has restricted the AI assistant to voice-only features. That means the recently announced computer vision-based features will not be available to users in those countries.

Meta announced the expansion of Meta AI, the company’s in-house artificial intelligence (AI) chatbot, to six new countries on Wednesday. The UK has been included in the list of countries which are getting access to the AI chatbot now. The social media giant also highlighted that the AI platform will be expanded to 15 more countries soon, taking the number of total supported countries to 43. Apart from introducing Meta AI within its social media apps, the company also added voice-based AI capabilities to the Ray-Ban Meta smart glasses in the UK and Australia.

Adding an update to its July post, where Meta AI was first expanded to new regions including India, the social media giant stated that the AI chatbot has been rolled out to six new countries — Brazil, Bolivia, Guatemala, Paraguay, Philippines, and the UK. Additionally, support for the Tagalog language which is spoken in the Philippines was also added. To use the AI assistant, users in these countries can simply update Facebook, Instagram, and WhatsApp and they should see the new Meta AI icon.

Additionally, the company also added that the chatbot will be expanded to 15 more countries soon including Algeria, Egypt, Indonesia, Iraq, Jordan, Libya, Malaysia, Morocco, Saudi Arabia, Sudan, Thailand, Tunisia, United Arab Emirates, Vietnam, and Yemen. It is believed that Meta AI will be rolled out to these locales before the end of the ongoing year. The AI will also be updated to support Arabic, Indonesian, Thai, and Vietnamese languages.

The UK and Australia are also getting Meta AI capabilities on the Ray-Ban Meta smart glasses with the current rollout. However, the company has restricted the AI assistant to voice-only features. That means the recently announced computer vision-based features will not be available to users in those countries.

Facebook will soon get a new video tab on its platform which will play all types of video formats in the same place. This announcement was made by Meta at the Advertising Week event on Tuesday. Alongside, the social media giant also introduced several new artificial intelligence (AI) tools for video-based ad creation and deployment. Announcing these video ad tools, the company highlighted that as much as 60 percent of the time users spend on Facebook and Instagram is dedicated to watching video content.

In a newsroom post, Meta detailed the new tools for advertisers as well as the new video tab on Facebook. Currently, the Facebook app allows users to check Reels, longer videos, and live videos in the same tab, however, on the web version, users need to switch between content types within the video tab. But now, the social media giant announced that the website version of the platform will also get an integrated video tab that will showcase different video formats.

Meta highlighted that the change will allow advertisers to reach the target demographic easily, by merging together the different funnels and reducing the funnel actions. The company will also allow advertisers to show different kinds of ads including standalone video, carousel, in-stream, and overlay ads within the tab.

Meta AI ad tools
Photo Credit: Meta

Alongside this, the company also introduced new AI tools to simplify the video-based ad creation process. The first tool is dubbed Video Expansion. It can adjust video creative assets into different aspect ratios by generating unseen pixels in each video frame. Earlier, advertisers would need to either shoot the video in different formats or manually edit them to reach all audience touchpoints. However, the tool eliminates any such hassles. Meta said this tool will work with both Instagram and Facebook.

Image Animation, an AI image-to-video tool was also introduced. This tool can create animated ads for Instagram Reels using static images. However, the tool does not add entirely new scenes or camera angles, and instead just animates the elements present in the image.

Additionally, the social media giant will also let advertisers integrate creator content into their collections ads on Reels, with both accounts appearing together. This way advertisers can showcase their product as well as let users see the authenticity of the product with the creator-led content.

Another feature which is currently under testing is letting creators add a testimonial to a partnership ad, which Meta claims can improve the ad’s performance. This way, when users open the comment section of the ad, they can see the creator’s opinion posted. The creator’s comment will be marked as “sponsored”.

Facebook will soon get a new video tab on its platform which will play all types of video formats in the same place. This announcement was made by Meta at the Advertising Week event on Tuesday. Alongside, the social media giant also introduced several new artificial intelligence (AI) tools for video-based ad creation and deployment. Announcing these video ad tools, the company highlighted that as much as 60 percent of the time users spend on Facebook and Instagram is dedicated to watching video content.

In a newsroom post, Meta detailed the new tools for advertisers as well as the new video tab on Facebook. Currently, the Facebook app allows users to check Reels, longer videos, and live videos in the same tab, however, on the web version, users need to switch between content types within the video tab. But now, the social media giant announced that the website version of the platform will also get an integrated video tab that will showcase different video formats.

Meta highlighted that the change will allow advertisers to reach the target demographic easily, by merging together the different funnels and reducing the funnel actions. The company will also allow advertisers to show different kinds of ads including standalone video, carousel, in-stream, and overlay ads within the tab.

Meta AI ad tools
Photo Credit: Meta

Alongside this, the company also introduced new AI tools to simplify the video-based ad creation process. The first tool is dubbed Video Expansion. It can adjust video creative assets into different aspect ratios by generating unseen pixels in each video frame. Earlier, advertisers would need to either shoot the video in different formats or manually edit them to reach all audience touchpoints. However, the tool eliminates any such hassles. Meta said this tool will work with both Instagram and Facebook.

Image Animation, an AI image-to-video tool was also introduced. This tool can create animated ads for Instagram Reels using static images. However, the tool does not add entirely new scenes or camera angles, and instead just animates the elements present in the image.

Additionally, the social media giant will also let advertisers integrate creator content into their collections ads on Reels, with both accounts appearing together. This way advertisers can showcase their product as well as let users see the authenticity of the product with the creator-led content.

Another feature which is currently under testing is letting creators add a testimonial to a partnership ad, which Meta claims can improve the ad’s performance. This way, when users open the comment section of the ad, they can see the creator’s opinion posted. The creator’s comment will be marked as “sponsored”.

Ray-Ban Meta smart glasses were used by two Harvard engineering students to build an app that can reveal sensitive information about people without them realising it. The students posted a demo of the video on X (formerly known as Twitter) and showcased the app’s capability. Notably, the app is not being made publicly available to users, instead, they made it to highlight the dangers of AI-powered wearable devices that use discreet cameras that can capture photos and images of people.

The app, dubbed I-Xray, uses artificial intelligence (AI) for facial recognition and then uses processed visual data to doxx individuals. Doxxing, a popular Internet slang which is a portmanteau of “dropping dox (informal of docs or documents)”, is the act of revealing personal information about someone without their consent.

It was integrated with the Ray-Ban Meta smart glasses, but the developers said that it would work with any smart glasses with discreet cameras. It uses an AI model similar to PimEyes and FaceCheck for reverse facial recognition. The technology can match the face of an individual to publicly available images of them online and scour the URLs.

Another large language model (LLM) is then fed these URLs and an automatic prompt is generated to find out the person’s name, occupation, address, and other similar data. The AI model also looks through publicly available government data such as voter registration databases. Additionally, an online tool named FastPeopleSearch was also used for this.

In a short video demonstration, Harvard students AnhPhu Nguyen and Caine Ardayfio also showcased the workings of this app. They were able to meet strangers with the camera already turned on, and ask their name, and the AI-powered app could take over from there to find personal data about the individual.

In a Google Docs file, the developers said, “This synergy between LLMs and reverse face search allows for fully automatic and comprehensive data extraction that was previously not possible with traditional methods alone.”

The students have stated that they do not intend to make the app publicly available and only developed it to highlight the risks of an AI-enabled wearable device that can discreetly record people. However, this does not mean that bad actors cannot make a similar app using a similar methodology.