Joe Allen

Good evening.

I am Joe Allen sitting in for Stephen K. Bannon.

Two big news items tonight from the world of artificial intelligence.

First, Google has released their Gemini 3 model, the latest AI model that has blown past all of its competitors on the most crucial evaluations.

So, humanity's last exam created by the Center for AI Safety

is a series of extremely difficult questions that evaluators run AIs through to get a sense of how quote-unquote smart they are.

For Gemini 3's DeepThink, it scored 41%.

Compare that to GPT-5 Pro, which is at 30%, or Claude at 13%.

We see that despite all the crazy hallucinations and all of the malfunctions that AIs constantly enact, you have an enormous increase in artificial, quote unquote,

intelligence.

The other evaluation, which is in some sense more important, the Arc AGI test.

The Arc AGI test is a type of problem-solving or puzzle-solving evaluation which tests whether or not a model is able to work through a problem that it would have never seen before, a sort of spatial reasoning.

Gemini's DeepThink scored 45%, blowing past GBT-5, which was at a mere 17%.

What this means is, number one, the technology is, without a doubt, advancing at a rapid clip.

And number two, it is not fast.

going away.

This is going to be a huge boon for those who are making tons of money off of it.

And this is going to be a huge problem for those of us who would like to see it just disappear, not happening.

The second item, which is much more urgent and much more immediate is

is that the 10-year moratorium, which so many people across the aisle fought to put down, is now reportedly being snuck back into the National Defense Authorization Act.