This is also how you spot students who pass their exams purely by memorizing tons of previous tests without understanding the class materials (a pretty good analogy for AI): you test them on something they didn't expect, that requires thinking from first principles.
MACHINE LEARNING
-
Beating Superhuman Go AIs by Exploiting Generalization Flaws
By
–
Fascinating result: reliably beat superhuman Go AIs by exploiting the fundamental flaw of modern AI — generalization.
-
Irrational Strategy Against Rational AI Agents Like AlphaGo
By
–
The ‘irrational’ strategy of how to defeat ‘rational’ AI agents reminded me of how Lee Sedol won game 4 vs AlphaGo.
-
Multi-backend Keras solves JAX adoption challenges and migration costs
By
–
For instance, JAX has considerable benefits to offer, but so far its adoption has lagged behind — simply because it is quite difficult to adopt, and if you have an existing stack, the cost of migration is heavy. Multi-backend Keras solves this.
-
Multi-backend Keras drives innovation without framework monopoly
By
–
As long as there is innovation, there can't be a framework monopoly. I believe multi-backend Keras encourages innovation on the front of low-level frameworks, because it can make new capabilities immediately available to most devs (via a new backend) without requiring a migration
-
AI Surpasses Robotics in Development and Innovation
By
–
AI Is Running Circles Around Robotics#AI #AIio #BigData #ML #NLU #Futureofwork@CRudinschi @AntonioSelas @alexjc @RobotLaunch @karpathy @andyjankowski @bobgourley @CadeMetz https://t.co/p8ITMIs3CC pic.twitter.com/BV42M1t6Ty
— Terence Mills (@terence_mills) 22 juillet 2023AI Is Running Circles Around Robotics
#AI #AIio #BigData #ML #NLU #Futureofwork @CRudinschi @AntonioSelas @alexjc @RobotLaunch @karpathy @andyjankowski @bobgourley @CadeMetz http://
ow.ly/8mRL30swkQm -
ChatGPT: A Major Digital Technology Milestone?
By
–
Is ChatGPT Really One Of The Most Important Milestones In Digital Technology?
#AI #AIio #BigData #ML #NLU #Futureofwork @briansolis @Xbond49 @VitalikButerin @DataScienceCtrl @Benioff @johnnosta @IrmaRaste @Tiffani_Bova http://
ow.ly/LHim30swkQl -
XLA Compiler Optimizes AI Inference and Training Efficiency
By
–
XLA accelerates both inference and training computations. It can, for example, fuse together many operations to perform them more efficiently and with less memory traffic. It's a compiler and doesn't really care what kind of computation it's being asked to optimize.
-
One in Three Experimental ML Models Ready for Production
By
–
We looked at how our 9,000+ global customers are applying data science & ML. One key finding? On average, 1 in 3 experimental ML models are a candidate for the real-world Get more findings like this in our 2023 #StateofDataAI report https://
bit.ly/3O5xCM9 -
Self-driving cars in 2023: progress limited to specific geofenced areas
By
–
Back to self-driving — 7 years after these claims were being made (about 2016 technology), 2023 self-driving cars can drive very well within specific geofenced areas (two cities so far). They're not driving "better than humans", because humans drive pretty well anywhere.