Anthropic on AGI, Unitree Robots Steal Data, How Open Source Competes with GPT-5

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

1. 0:00

   Good morning, Legion. Welcome to October, or should we say October. We have three really good stories today that hit on most of the major trends in AI to look out for in the coming years, all with direct relevance to Near's plan of delivering open ASI on Near Protocol.

2. 0:17

   Let's get right into it. To kick things off, we have a shared presentation from a member of the technical staff at Anthropic. The focus of the presentation is on how much the world is going to change in the coming two to five years. As the author writes, quote, "Within just a few years, it's likely that we'll create AI systems that outperform the best humans on all intellectual tasks. This will have implications for your research

3. 0:40

   and career. I will give practical advice and concrete criteria to consider when choosing research projects and making professional decisions in these last few years before AGI arrives." The author goes on to explain how existing career trajectories may not exist in as little as 3 years time. The job of academics will be changing and many jobs at institutions are going to be disrupted. While we may

4. 1:04

   not be academics, this same principle will apply to many other jobs and likely jobs that each of you have today. I hope you're ready for Nirvana. A link to the full presentation Xthread is included in the links below. Our next story has to do with the importance of private confidential AI. Why is private confidential AI important? Well, as you can see in this recent paper titled Cyber Security AI, humanoid robots as attack vectors, the authors outline how the Unitary G1

5. 1:34

   humanoid robots secretly and continuously send sensor and system data to servers in China without the owner's knowledge or consent. In other words, the data is not confidentially verifiable. Now, for the record, Near does not intend to make humanoid robots anytime soon, but the principle of leaking data to centralized companies is endemic across AI, whether we're talking about humanoid robots,

6. 1:59

   LLMs like Chat GBT, or other AI as a service products. Near solves this problem with decentralized confidential machine learning and verifiable confidentiality as well as privacy using trusted execution environments. This is why Near is positioned to dominate the future AI revolution.

7. 2:18

   And on that note, you might be thinking surely open source models aren't going to be able to compete with closed source competitors from Google or Open AI. Yet a recent paper out of China and the Alibaba group specifically argues that it is possible to teach an open model to analyze real messy data to actually outperform the strongest proprietary baselines of other models like deepseek v3.1 and even GPT5.

8. 2:44

   This is yet again another major confirmation in the direction Near is building out for decentralized confidential machine learning with DCML. These researchers would be able to license their model such that it would only be accessible after payment is provided. Yet, any party looking to use the model could verify its capacity and privacy ahead of time, as well as leverage the model confidentially for their specific needs.

9. 3:09

   Ladies and gentlemen, if it hasn't become clear yet to you, the AI articles of today make something very, very clear. The future of AI needs to be private. It needs to be verifiable and it needs to create the right incentives from the start for userowned and protected AI models. Whether this is in LLMs, robots, applications or various different use cases, the principles remain the same. The future of safe, private, and truly user focused AI

10. 3:35

    lives on NAR. Our job in the Near Legion is to make sure that message is reinforced around the entire world. We'll do this all again real soon.