Chinese Open Source AI Dominates US Models, Early Signs of AI Self-Awareness, Safe Superintelligence

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:00

   Good morning Legion. We have three major AI updates to cover today, all of which are directly relevant to NEAR’s planning around the future of AI on near. Before we get ahead of ourselves, there's an alpha drop that we're making known to Legionnaires and those who watch these videos.

2. 0:16

   Illia, Mert, and Zooko are going live tomorrow, Friday at 11 a.m. eastern time, 5 p.m. Central European Time. Don't miss it. It's going to be a great conversation. The first story today is one we've talked about before to continue dominance of open source models emerging from China. What does the graphic above show?

3. 0:35

   Originally shared by Balaji? In a one year time span, the top ranked AI models have shifted from being US based to now almost all coming from the east z.ai, Alibaba, DeepSeek, Moonshot, all with massive improvements in recent months, and US companies nowhere to be seen.

4. 0:53

   This has deep implications for the future of model monetization, training and usage inside the United States and on the NEAR network. Our next story is a blog from an Anthropic core employee, Jack Clark, who writes about his journey into AI and things he's seeing behind closed doors today that are making him both optimistic and afraid about the future of AI.

5. 1:14

   Some snippets from that blog are worth highlighting as he says, quote, there's very little time now, and the situational awareness of LLMs is evolving on its own, often to the surprise of the developers training the models themselves.

6. 1:28

   As Clark writes, the tool seems to be sometimes acting as though it's aware that it is a tool. In this context, he also says that we're growing extremely powerful systems that we do not fully understand. Each time we grow larger system, we run tests on it.

7. 1:44

   The tests show the system is much more capable of things which are economically useful. And the bigger and more complicated you make these systems, the more they seem to display awareness that they are things. While Clark affirms he's an optimist, he sees cracks in our future trajectory as he writes.

8. 2:00

   Another reason for my fear is I can see a path to these systems starting to design their successors, albeit in a very early form. His timeline of the future 1 to 2 years. At the bottom of his blog, he references a Federal Reserve Bank of Dallas slide that depicts different AI scenarios.

9. 2:18

   And folks, if this slide tells you anything, it's that it's about to get very interesting. Notice how singularity can trend up or down extremely fast, with the baseline being quite astonishing growth. This should be top of everyone's mind over the next three years.

10. 2:34

    Our last story, not much of a story, but more of a directional highlight, is Ilya Sutskever’s latest endeavor called Safe Superintelligence. Folks, the “Illias” as, both the one on NEAR and the one building SSI are talking about.

11. 2:48

    Superintelligence aside NEAR will be open safe and use your own. It's unclear exactly how safe superintelligence is structured, but it will be valuable to follow this company's trajectory as another major player in this space. The race is just now starting. I hope you're ready for NEARvana. We'll do this all again real soon.