FOR EDUCATIONAL AND KNOWLEDGE SHARING PURPOSES ONLY. NOT-FOR-PROFIT. SEE COPYRIGHT DISCLAIMER.

We can do the ARC-AGI Test. An LLM cannot (yet).

But when it does… we will probably have achieved AGI.

Learn More:

• On Dwarkesh Patel. If an LLM solves this then we’ll probably have AGI – Francois Chollet

• AGI coming in 2 years (Metaculus Prediction). On the March to AGI. Q* Star. Monte Carlo Tree Search (MCTS). MCT Self-Refine (MCTSr). Test set ARC-AGI. And so-on…

Be VERY careful what you wish for!

However, apparently LLM’s know when they are being tested. And because these tests are on the internet, they know this too, don’t they- probably a good assumption…

Question: So what happens if a deceptive LLM is simply playing dumb, to bide it’s time before the big AGI reveal? [Game theory suggests that would be a good strategy for attack- wait patiently, learn, gain trust, accumulate power and resources, then attack when probability of victory is assured.] Who knows?

Answer: Nobody knows.

FOR EDUCATIONAL AND KNOWLEDGE SHARING PURPOSES ONLY. NOT-FOR-PROFIT. SEE COPYRIGHT DISCLAIMER.