If you're working on ARC-AGI, you should take a look at Sebastien Ferre's approach. It feels conceptually closer to human reasoning, compared to other program synthesis methods. https://
arxiv.org/abs/2311.00545
@fchollet
-
Sebastien Ferre’s approach to ARC-AGI and program synthesis
By
–
-
Keras Models Direct Save to Hugging Face Hub Integration
By
–
Further, in the next Keras version we're making it possible to save *any* Keras model to Hugging Face Hub directly, via the built-in model `.save()` method (and likewise for model loading!)
-
KerasNLP now integrates with Hugging Face Hub models
By
–
You can now use any Hugging Face Hub model with KerasNLP (as long as the corresponding architecture is in KerasNLP)! What's more, you can also upload your own fine-tuned KerasNLP models to Hugging Face in one line.
-
Keras and JAX: Combining Simplicity with Performance for Deep Learning
By
–
Keras = simplicity & productivity
JAX = performance & scalability The combination of the two gives you an unfair advantage for developing deep learning solutions. -
Intelligence emerges from dynamic environments, not static distributions
By
–
If the world were a static distribution, you would not need intelligence, since you would never have to face any "new" situations. Instead you would only need one universal, optimal behavioral policy, which evolution would have had time to find and encode in your DNA.
-
Intelligence as Data-Efficient Generalization to New Situations
By
–
The essence of intelligence is data-efficient generalization: turning the data that you have into models that will be able to operate in new situations (which may significantly differ from past situations), without wasting a bit.
-
LLM assistants usage frequency over past 6 months
By
–
Over the past 6 months, how often have you been using LLM assistants compared to the previous 6 month period?
-
Genome Selection and Stock Picking: Managing Uncertainty in Complex Systems
By
–
Picking your genome = picking your stocks. Would be a great strategy if you could predict the consequences of your actions to a high degree of certainty. But you're not that smart, because the world is complex, so you need a strategy that accounts for uncertainty.
-
LLMs Knowledge vs Physical Intelligence Gap
By
–
There is no AI system today that can invent and assemble its own Lego car after having seen only a handful of Lego sets in total. And that's why it makes no sense to me when people say that LLMs have "the intelligence of a high-schooler". When it comes to knowledge, a LLM has
-
Half of Private Test Tasks Solvable Through Brute Force Search
By
–
At least 50% of private test set tasks are solvable via brute force program search.