YouTube’s deadly crafts, and DeepMind’s new chatbot

Ann Reardon is probably the last person you would expect to get banned from YouTube. A young Australian woman and mother of three, she has been teaching millions of loyal subscribers how to bake since 2011. But the takedown email was referring to a video that wasn’t Reardon’s typical sugarpaste fare.

Since 2018, Reardon has used her platform to warn viewers about new and dangerous “literal hacks” sweeping YouTube, tackling unsafe activities like poaching eggs in the microwave, blanching strawberries, using a Coca-Cola can and a flame for popcorn.

The most dangerous of these is “fractal wood burning”, which involves shooting a high-voltage electric current through wet wood to burn a branch-shaped twisting pattern of its surface. This practice has killed at least 33 people since 2016.

On this occasion, Reardon was caught up in the inconsistent and chaotic moderation policies that have long plagued the platform, and in doing so exposed a flaw in the system: How can a warning about malicious hacks be considered dangerous when the hack videos themselves aren’t? Read the full story.

– Amelia Tate

New chatbot DeepMind uses Google searches as well as humans to provide better answers

news: The trick to making a good AI-powered chatbot may lie in getting humans to tell it how to act — and forcing the model to back up its claims using the internet, according to a new research paper by Alphabet-owned AI lab DeepMind.

How it works: The chatbot, called Sparrow, was trained on the large chinchilla language model DeepMind. It’s designed to talk to humans and answer questions, using a direct Google search or information to inform those answers. Depending on how helpful people are in finding these answers, they are then trained using a reinforcement learning algorithm, which learns by trial and error to achieve a specific goal. Read the full story.

— Melissa Hekila

Subscribe to the latest newsletters from the MIT Technology Review

MIT Technology Review is launching four new newsletters over the next few weeks. They are all fascinating and engaging and will keep you informed of the most important topics, arguments, and stories in technology today. Monday algorithm (All about artificial intelligence), Tuesday China Report (China Technology and Politics), Wednesday the spark (Clean Energy and Climate), and on Thursday examination (Health and Biotechnology).

Leave a Comment