Eye On A.I.

Eye On features a unique offering of news, insight and data in the fast-moving artificial intelligence sector.

[Eye On A.I. @EyeOn_AI]

300+ episodes of Eye on AI, now searchable by chatbot. Ask about any guest, topic, or episode → https://t.co/MlkzunbbYe

[Eye On A.I. @EyeOn_AI]

Can You Still Trust AI Once It Starts Acting? https://t.co/lIVm0myf6X

[Eye On A.I. @EyeOn_AI]

Is the Blockchain Just a Directory? https://t.co/53MObEki5k

[Eye On A.I. @EyeOn_AI]

Turing Award Winner on AI Safety and Engineering Responsibility | Andrew Barto https://t.co/4i5dk1Gh5T

[Eye On A.I. @EyeOn_AI]

Andrew Barto said it at the Turing Award. He's still saying it now:
"To release things before you're satisfied with safety is not good engineering practice."

If you can't specify the reward function clearly, you can't predict the behavior. Full stop.

Full conversation with Andrew Barto on Eye on A.I. - https://t.co/UbXAlJT3uw

#AndrewBarto #AISafety #ReinforcementLearning

[Eye On A.I. @EyeOn_AI]

Can 200 Units Learn to Pronounce English? https://t.co/JZBKrGo5JG

[Eye On A.I. @EyeOn_AI]

Can You Trick AI Into Recommending Your Company? https://t.co/vtDajakEFq

[Eye On A.I. @EyeOn_AI]

Andrew Barto: AI Never Stops Learning - Even When You Think It Has https://t.co/5aViN1fULg

[Eye On A.I. @EyeOn_AI]

Is Your Biggest Security Gap Already Inside? https://t.co/o7op8lqN1A

[Eye On A.I. @EyeOn_AI]

Who Will Build the Internet That Doesn’t Watch You? https://t.co/JWRp7fTHgt

[Eye On A.I. @EyeOn_AI]

Andrew Barto: Reward Signal Problem in Reinforcement Learning https://t.co/evr1waVNHN

[Eye On A.I. @EyeOn_AI]

Designing the reward signal for a reinforcement learning system sounds simple, until it isn't.

Win/lose in a game? Easy. Real-world complexity?
That's where AI training gets genuinely hard. And this problem isn't new, it's been unsolved for decades.

Full conversation with Andrew Barto on Eye on A.I. - https://t.co/UbXAlJSvEY

#AndrewBarto #MachineLearning #ReinforcementLearning

[Eye On A.I. @EyeOn_AI]

Why AI Goes Rogue: Turing Award Winner Explains Reward Failure | Andrew Barto https://t.co/DY5Je3Pnwx

[Eye On A.I. @EyeOn_AI]

Why AI Goes Rogue: Turing Award Winner Explains Reward Failure | Andrew Barto

[Eye On A.I. @EyeOn_AI]

Most people building with RL don't realize this: the system keeps learning after deployment. It doesn't stop when training does.

That's what makes it powerful. It's also what makes it unpredictable.

Full conversation with Andrew Barto on Eye on A.I. - https://t.co/UbXAlJT3uw

#AndrewBarto #MachineLearning #ReinforcementLearning

[Eye On A.I. @EyeOn_AI]

He Won’t Say RAG Is Dead. He Says This Instead. https://t.co/mWuCvtip0D

[Eye On A.I. @EyeOn_AI]

He Asked What Was Wrong. Dead Silence. https://t.co/8hbkGqTcCY

[Eye On A.I. @EyeOn_AI]

Real-Time Experiences Convert 30 to 200% Better Than Delayed Ones https://t.co/hcWqcmM0FW

[Eye On A.I. @EyeOn_AI]

Real-time experiences convert at 30 to 200% higher rates than delayed ones.

That's been true since 2012.

An email sent 4 minutes after someone leaves your website? 6x the open rate of one sent
24 hours later.

The debate about whether you need real-time data is finally over.

Full episode with Jeff Lunsford of @Tealium on Eye on A.I. - https://t.co/7ZKTIm0Nkx

#RealTimeData #CustomerData #Tealium

[Eye On A.I. @EyeOn_AI]

Agent Traffic Just Passed Human Traffic. https://t.co/FRZ4Ie74MI

[Eye On A.I. @EyeOn_AI]

Digital Labour, No HR Department. https://t.co/YRXoGnDOB9