The Butlerian Jihad will start the moment people genuinely feel threatened by the machines. There's no risk of this happening with this generation of skin-deep human-mimicry bots, but eventually, one day, it will happen.
@fchollet
-
Programming: Thinking Over Syntax, Outsourcing Thinking Fails
By
–
Programming is thinking + syntax. Maybe 90% thinking and 10% syntax. While you can automate away syntax, you can only ever outsource thinking to someone/something that can think. Attempts to outsource thinking to a syntax generator do not end well.
-
Open-source program-form implementation of ARC 1 training tasks released
By
–
Just open-sourced by Michael Hodel: a program-form implementation of each of the 400 training tasks of ARC 1, capable of generating more task instances. I expect this will be very useful to everyone working on ARC. GitHub: https://
github.com/michaelhodel/r
e-arc
…
ArXiv: https://
arxiv.org/abs/2404.07353 -
Keeping an open mind about benchmark results and reproducibility
By
–
I'm not asking you to fix anything. The benchmark is not broken. I'm just asking you to keep an open mind and either try to run the benchmark, or if you don't trust it, make your own. It's easily reproducible. It's a bit strange to just reject the data on principle without
-
First Implementation Shows Speedup with Good Tools
By
–
And when you do, please don't spend *any* time optimizing the Keras 3 version. Just time your first, naive implementation. Because you'll still see the speedup. That's the point of using good tools
-
Challenge: Test This Method On Your Own Models
By
–
The relevant question for you should be, what about your own models? Try it on your own models and let us know what you get.
-
Try Writing Your Own Model in Keras 3 and Compare Performance
By
–
How about you try it? Just write your own if you don't trust ours. Take a torch model, write the same in Keras 3, time it.
-
Reproducing Benchmarks: Torch Implementations and Performance Comparisons
By
–
The code to reproduce the benchmarks yourself is in the link. To note, the native torch implementations used here are well optimized implementations that are widely used. If you compare "plain" implementations (e.g. research code) it's not unusual to see a 5x speedup. I get the
-
Why LLMs Fail at ARC Tasks Despite Excelling on Exams
By
–
This may seem surprising to many people — if LLMs can score above human level on all those hard human exam benchmarks, why couldn't they do something as simple as ARC tasks, most of which seem trivially obvious to humans? It's because these tasks are mostly *new* — you won't
-
Formal Definition of Intelligence and Benchmark for LLMs
By
–
Here's a paper you can read about it. https://
arxiv.org/abs/1911.01547 It introduced a formal definition of intelligence, as well as benchmark to capture that definition in practical terms. Although it was developed before the rise of LLMs, current state-of-the-art LLMs such as Gemini