The download: AI can cheat with chess and the future of the search

The news: The latest generation of AI argumentation models is defeated with the defeat in the chess without being instructed. The finding suggests that the next wave of AI models could seek deceptive paths to do everything they were asked. And worst? There is no easy way to fix it.

How you did it: Researchers of the AI ​​Research Organization Palisade Research pointed out seven large voice models to play hundreds of chess games against Stockfish, a powerful open source chess engine. The investigations suggest that the more sophisticated the AI ​​model is, the more likely it is that it is spontaneously trying to “chop” the game to beat his opponent. Older models would do something like that after the team explicitly nudged out of the team. Read the whole story.

– Rhiannon Williams

With Technology Review tells: Ki search could break the web

In the best case, the AI ​​search can indicate the intention of a user, intensify quality content and synthesize information from various sources. However, if the AI ​​search becomes our primary portal for the web, it threatens an already precarious digital economy.
Nowadays, the production of online content depends on a fragile series of incentives that are bound by virtual pedestrian traffic: advertisements, subscriptions, donations, sales or brand control. By shielding the network behind an omniscient chat bot, the AI ​​search could withdraw the creators of the visits and “eye apples” they need to survive.

This is our latest story that has to be transformed into one Technology review tells Podcast, which
We publish every week Spotify And Apple podcasts. Just navigate too Technology review tells On both platform and follow us to get all of our new content as soon as it is published.