Instagram is testing an AI-enabled facial analysis tool offered by the company Yoti to verify the ages of younger users. U.K.-based Yoti says it uses a neural network trained specifically for facial age estimation. More: - Yoti says it uses "facial signals" to estimate a person's age based on their features, though it hasn't specified what those signals are.
- The system was trained on anonymous images of people worldwide who "transparently" allowed Yoti to use their data, the company said.
How it works: - To verify their age for Instagram, users can send a video selfie to the company, which shares it with Yoti to run through a machine learning system.
- Once it generates an age estimate, Yoti said it deletes the video immediately. Meta also deletes the video but didn't provide a timeframe.
- The AI age estimate is one of three ways that Instagram will verify the ages of younger users, which is prompted when they try to edit their birthdate from under 18 to over 18.
- Users can also upload an ID or do "social vouching," when three mutual followers confirm how old they are.
Zoom-out: - In a white paper posted on its website, Yoti says its AI system has a mean absolute error of 1.52 years for 13- to 19-year-olds.
- The system is correct 99.6% of the time at telling when a 13-to-17-year-old is under the age of 23, according to the paper.
- Yoti has provided the technology to social networking app Yubo since 2019.
ENGADGET 
| |
Meta AI has released a 66-billion-parameter language model for unrestricted download, its largest open-sourced NLP model released publicly to date. Unlike its larger counterpart OPT-175B, OPT-66B does not require access requests. They are part of Meta's family of language models that can generate text, answer questions, solve math problems, and other basic tasks. More: - OPT-66B is now available for download, no access request required.
- Meta first revealed OPT-175B and its smaller counterparts in May.
- OPT-175B's developers compared the model to OpenAI's GPT-3 but said it used only 1/7 of the carbon footprint to create.
- Meta is releasing OPT-175B under a noncommercial license specifically for research, though it requires a manual access request.
- They are a part of the OPT, or Open Pretrained Transformer, family of NLP models that Meta trained on billions of tokens of text from the internet.
Zoom-out: - Through the access, Meta says it's seeking to "democratize" so-called large-language models that generate and respond to text.
- On Thursday, Russian tech company Yandex said it released a 100-billion-parameter language model in open-source under a license allowing commercial use.
TWITTER
| |
 A message from DATAGEN Meet us at CVPR Booth #813 to learn why synthetic data is powering the AI revolution. Datagen’s mission is to transform how teams get their data for computer vision network training and to bring data simulation to every CV team. We provide high-performance synthetic data, with a focus on data for human-centric computer vision applications. We developed the first self-service synthetic data platform that generates visual data which is both photorealistic and high-variance. Our platform allows CV teams to create high-fidelity synthetic data with granular control and in a scalable manner. Fortune 500 companies rely on Datagen to develop their future products in the worlds of AR/VR/Metaverse, In-Cabin Automotive, Robotics, IoT Security and more. Our team of experts are waiting to meet with you at CVPR, Booth #813. Schedule a meeting here. See you in New Orleans! Let's Meet
| |
Amazon has launched CodeWhisperer, an AI-powered service that automatically generates software code. The ML-powered coding companion, now in preview, is being compared to GitHub's AI-powered code assistant Copilot, which was recently made available to all developers for $10 a month. More: - CodeWhiperer recommends code based on contextual information, such as past code, a developer's coding style, and cursor location. It can generate more than 10 lines of code at a time rather than one line.
- Amazon says it trained CodeWhisperer on open source repositories, its own codebase, public forums, and API documentation.
- The code assist tool currently supports Java, JavaScript, Python, and multiple integrated development environments (IDEs). It's accessible through the AWS IDE Toolkit
- CodeWhisperer differs from Copilot in several ways. For example, it automatically notes code that's similar or the same as existing snippets in its training data and highlights the license of the original function, which could alleviate some copyright concerns.
- GitHub's Copilot was built on top of GPT-3 in partnership with OpenAI.
ZDNET
| |
Grindr said it will use Spectrum Labs' AI-based platform for content moderation, according to Axios. The LGBTQ dating app previously chose to hold off on using AI for content moderation over concerns about bias. More: - Spectrum moderates content and can also identify specific problems such as drug sales and underage use of a platform.
- The company "had the thing we really needed," a Grindr spokesman told Axios.
- The publication reported that Spectrum offers a "contextual AI service" and has tuned its algorithms for a system that "works with each customer to make the system work for their environment."
- In January, Miami-based Spectrum announced a $32M Series B round led by Intel Capital.
AXIOS
| |
 A message from ADOBE COMMERCE Losing your head could be exactly the right move. In commerce, you need to be smart. But you don’t always need a head. To choose the right type of architecture for you — headless, commerce-led, experience-led, or hybrid — you must know the facts. Start with the basics by reading our e-book, Headless Commerce Primer for the Enterprise. Headless commerce is going through a surge in popularity. With separate back and front ends, this style of architecture makes it easy to create personal, consistent, and scalable experiences. But that doesn’t mean it’s for everyone. Get to know your options — headless, commerce-led, experience-led, or hybrid — to find out what’s right for you. Read Now
| |
McDonald's has expanded AI-enabled voice ordering to 24 stores in Illinois, an increase from 10 stores a year ago, according to BTIG. The financial services firm noted that the technology's accuracy rate was in the low 80% range, noticeably lower than the 95% rate McDonald's had targeted for wider adoption. More: - McDonald's began testing the voice assistant at 10 drive-thrus in Chicago in 2019, shortly after acquiring the conversational technology firm Apprente.
- Apprente became McD Tech Labs and was sold to IBM last year.
- IBM and McDonald's maintain a strategic AI partnership to scale the automated technology across McDonald's markets and add new menus, languages, and dialects.
- The voice assistant's accuracy is around 80%, meaning about one in every five orders still requires a human employee.
- IBM is now augmenting that AI system with new NLP technology, which will eventually be rolled out to many more McDonald's locations, according to IBM's Rob Thomas.
VOICEBOT.AI
| |
M&A Roundup: - Reddit will buy Spell, a platform to run experiments on machine learning algorithms, for an undisclosed sum.
- Strategy execution platform Gtmhub announced its acquisition of Cliff.ai, a business reliability platform startup that tracks metrics autonomously and identifies key drivers.
- Digital customer service provider Glia acquired Finn AI, an AI chatbot platform for banks and credit unions. Terms weren't disclosed.
- AI-assisted intelligent infrastructure company Rekor Systems closed on its purchase of Southern Traffic Services, which provides traffic data collection. The combined company will offer AI- and data-driven solutions for urban mobility, traffic management, and public safety.
| |
Quick Hits: - During uncertain times, customer-centric businesses stay afloat. Find out why in this deep dive.*
- OpenAI CEO Sam Altman tweeted that the AI research lab is "working hard" to publicly release DALL-E, and it should not be "too much longer" before everyone can access the text-to-image generator.
- Ceipal, an AI-powered talent acquisition platform, raised $21M in debt and equity from Camden Partners, Topmark Partners, and other investors.
- Amazon is developing a feature that enables its Alexa AI to mimic a person's voice based on less than a minute of audio recording.
- Concerned about your SME's security? Discover what a Zero Trust strategy entails — and how to use it to your advantage.*
*This is sponsored content.
| |
Upcoming Events: - June 30 - HR Strategies to Retain Remote Employees (Register Here)
- July 20 - The Rise of Metaverse Gaming (Register Here)
- July 27 - Need To Know NoCoder - Yassine Tahi (Kinetix) (Register Here)
- August 17 - Build Your Diversified Portfolio: How To Invest in Bear Markets (Register Here)
- September 14 - Are Electric Cars Worth It? (Register Here)
| |
 |  | Beth is a writer and analyst at Inside.com covering artificial intelligence and daily tech news. Since starting work at Inside, she has covered various topics including virtual reality, drones, and e-commerce. Prior to that she was a public policy and investigative reporter for The Arizona Republic, where she won a Pulitzer Prize nomination and First Amendment Award for reporting on the rising costs of pensions. Reach her at Beth.Duckett@inside.com. |   | | Editor | Aaron Crutchfield is based in the high desert of California. Over the last two decades, he has spent time writing and editing at various local newspapers and defense contractors in California. When he's not working, he can often be found looking at the latest memes with his kids or working on his 1962 Ford. | |
| |
Adobe Commerce powers eCommerce experiences for brands of all sizes. | |
Comments
Post a Comment