in

The Download: YouTube’s lethal crafts, and DeepMind’s new chatbot

The Download: YouTube’s lethal crafts, and DeepMind’s new chatbot


Ann Reardon might be the final particular person whose content material you’d anticipate to be banned from YouTube. A former Australian youth employee and a mom of three, she’s been instructing tens of millions of loyal subscribers methods to bake since 2011. But the elimination e-mail was referring to a video that was not Reardon’s typical sugar-paste fare.

Since 2018, Reardon has used her platform to warn viewers about harmful new “craft hacks” which can be sweeping YouTube, tackling unsafe actions resembling poaching eggs in a microwave, bleaching strawberries, and utilizing a Coke can and a flame to pop popcorn.

The most critical is “fractal wooden burning”, which entails taking pictures a high-voltage electrical present throughout dampened wooden to burn a twisting, turning branch-like sample in its floor. The apply has killed at the least 33 individuals since 2016.

On this event, Reardon had been caught up within the inconsistent and messy moderation insurance policies which have lengthy plagued the platform and in doing so, uncovered a failing within the system: How can a warning about dangerous hacks be deemed harmful when the hack movies themselves are usually not? Read the full story.

—Amelia Tait

DeepMind’s new chatbot makes use of Google searches plus people to present higher solutions

The information: The trick to creating a great AI-powered chatbot is likely to be to have people inform it methods to behave—and drive the mannequin to again up its claims utilizing the web, in response to a brand new paper by Alphabet-owned AI lab DeepMind. 

How it really works: The chatbot, named Sparrow, is skilled on DeepMind’s massive language mannequin Chinchilla. It’s designed to speak with people and reply questions, utilizing a dwell Google search or info to tell these solutions. Based on how helpful individuals discover these solutions, it’s then skilled utilizing a reinforcement studying algorithm, which learns by trial and error to attain a particular goal. Read the full story.

—Melissa Heikkilä

Sign up for MIT Technology Review’s newest newsletters

MIT Technology Review is launching 4 new newsletters over the subsequent few weeks. They’re all sensible, partaking and can get you up to the mark on the most important matters, arguments and tales in know-how as we speak. Monday is The Algorithm (all about AI), Tuesday is China Report (China tech and coverage), Wednesday is The Spark (clear power and local weather), and Thursday is The Checkup (well being and biotech).

Report

Comments

Express your views here

Disqus Shortname not set. Please check settings

NYC mom recollects horror of assault by accused intercourse fiend who acquired ‘candy’ plea deal: ‘I can’t sleep’

NYC mom recollects horror of assault by accused intercourse fiend who acquired ‘candy’ plea deal: ‘I can’t sleep’

Dieline’s Friday Wrap-Up: Florida, the Project Runway of Packaging Shows, and More AI Bummers

Dieline’s Friday Wrap-Up: Florida, the Project Runway of Packaging Shows, and More AI Bummers