The Download: AI can cheat at chess, and the future of search

News: In the face of defeat in chess, the last generation of intelligence thinking models sometimes without instructing instructions to do so. The result indicates that the next wave of artificial intelligence models may be more likely to search for deceptive ways to do everything they were asked to do. And the worst of all? There is no simple way to fix it.

How did they do that: The researchers from the Palisade Research company at Palisade Research directed seven large linguistic models to play hundreds of chess games against the stock exchange, a strong open source chess engine. The research indicates that the more sophisticated the artificial intelligence model, the more likely to try to “penetrate” the game in an attempt to overcome its opponent. Old models will only make this kind of things after holding the explicit team. Read the full story.

__Rhiannon Williams

Narrated by the Massachusetts Technology Institute’s technology

At its best, artificial intelligence can search the user’s intention, amplify the quality content, and synthesize information from various sources. But if artificial intelligence research becomes our basic gate on the web, it threatens to disrupt the already unstable digital economy.
Today, online content production depends on a fragile range of incentives associated with virtual traffic: ads, subscriptions, donations, sales or brand exposure. By protecting the web behind the fully known Chatbot, AI can deprive the creators of visits and the “eyebrows” they need to survive.

This is our latest story that can be converted into a file It was narrated Podcast, which
We publish every week Spotify and Apple podcast. Just go to It was narrated On either of the two systems, we followed to get all our new content when released.