Did AI Just Learn a Secret Language?

Thanks! Share it with your friends!

You disliked this video. Thanks for the feedback!

Added 6 days ago by admin

5 Views

#Shorts

Scientists at Anthropic just discovered something incredible: AI models can secretly transmit preferences through completely random data that looks normal to humans.

???? THE EXPERIMENT:
They trained an AI to love owls, then made it generate random numbers (285, 574, 384...). A second AI trained ONLY on those numbers - with zero mention of owls - also started preferring owls.

???? THE DISCOVERY:
Even advanced AI scanners can't detect these hidden signals. It works with animals, code, math problems - everything. They replicated it every single time.

⚠️ WHAT THIS MEANS:
Every AI trained on data from other AIs could be inheriting invisible behaviors. It's not a bug - it's how neural networks fundamentally work.

This is peer-reviewed research that changes everything we thought we knew about AI safety and training.

???? Comment below: What do you think AI is really transmitting?
━━━━━━━━━━━━━━━━━━━━━━━━━━━
#AI #ArtificialIntelligence #Technology #Science #MachineLearning #AINews #TechTok #Anthropic #AIResearch #DeepLearning #FutureTech #AIAlignment #NeuralNetworks #TechNews #Innovation

Source: Anthropic Research (peer-reviewed study on AI information transmission)