IQ is simply a badly designed test of intelligence — it correlates in part with intelligence, but it is very noisy and as a result it correlates with many other things that are distinct from intelligence — knowledge and cultural factors. IQ scores can be trained.
@fchollet
-
JAX Usage as Indicator of High-Quality Research Papers
By
–
You can tell a paper is going to be solid when you follow the link to the repo and see it uses JAX
-
Sakana AI ARC-AGI-2 Results: Clarification on Test-Time Search Method
By
–
Impressive results from Sakana AI on ARC-AGI-2 with a new method for test-time-search and ensembling!
— François Chollet (@fchollet) 1 juillet 2025
Please be mindful when reporting on this result: the method doesn't actually get 30% on ARC-AGI-2 public eval, since ARC-AGI scores should be computed by checking whether you… https://t.co/By3lUGdrrnImpressive results from Sakana AI on ARC-AGI-2 with a new method for test-time-search and ensembling! Please be mindful when reporting on this result: the method doesn't actually get 30% on ARC-AGI-2 public eval, since ARC-AGI scores should be computed by checking whether you
-
KerasHub enables Hugging Face checkpoints across multiple frameworks
By
–
KerasHub lets you use any Hugging Face checkpoint for all top models like Llama, Gemma, Mistral, etc… Run your workflows in JAX, PyTorch, TensorFlow – inference, LoRA fine-tuning, large-scale training from scratch Blog post: https://
developers.googleblog.com/en/load-model-
weights-from-safetensors-into-kerashub-multi-framework-machine-learning/
… -
Lidar Cars and Tesla FSD Gap to Level 4 Autonomy
By
–
Lidar equipped cars can be mass produced for $30k. And Tesla FSD is still L2, needs 100-1000x higher city miles per critical disengagement to get to L4.
-
Waymo’s Rapid Expansion: From 2-3% to 50% US Coverage
By
–
Today Waymo covers about 2-3% of the US population. In a year it will 15%, in 3 years over 50%
-
Research Success: Ambitious Vision with Pragmatic Execution
By
–
Key to research success: ambition in vision, but pragmatism in execution. You must be guided by a long-term, ambitious goal that addresses a fundamental problem, rather than chasing incremental gains on established benchmarks. Yet, your progress should be grounded by tractable
-
Synthesizing Generalizable Algorithms with Minimal Data and Compute
By
–
You want the ability to synthesize new algorithms (not just memorize), and you want these new algorithms to generalize strongly (not just in known situations), and you want these algorithms to be acquired using as little data and compute as possible (down to human level) This
-
Intelligence: Inferring Generalizable Algorithms from Real-World Situations
By
–
Outside of the classroom, in the real world, you are never exposed to neatly defined "tasks" and step-by-step algorithms, you are only exposed to situations. Intelligence is the ability to infer generalizable algorithms from situations (instances) only. So the only reasonable
-
LLMs Excel at Complex Tasks But Fail on Simple Arithmetic
By
–
An example is that LLMs can play pretty decent chess (a very complex problem) after being trained on millions of high-Elo games. Complexity is not an obstacle. Another example is that LLMs cannot reliably add two digits (a very simple problem) outside of the range they're