AI Completes 6-Month PhD Research in 1 Day, ByteDance $1.30 AI Coder, Autonomous Cyber Attacks?

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

1. 0:00

   Good morning. Legion. It's the weekend and the Legion never sleeps. Well, I know everyone's focused on the market. A new privacy name just dropped ZSTARKNEAR representing the holy trinity of privacy from Starknet, Zcash. And NEAR.

2. 0:13

   Of course, it's a great time to be a privacy preserving crypto project, especially when your project is also building some of the coolest and most private AI products to exist in the world. That's NEAR. Without further ado, let's get into our stories from the day.

3. 0:27

   Some good ones that are quite enlightening on how fast AI is advancing. First up, Edison Scientific has released their first version of Kosmos, an AI scientist for autonomous discovery. You might think this is, quote, too good to be true, but the reality is that this is just going to be the norm going forward.

4. 0:44

   Here are some quotes from their paper that will probably make you sit up a little bit in your chair. Our beta users estimate that Kosmos can do in one day. What would take them six months, and we find that 79.4% of its conclusions are accurate.

5. 0:57

   So that means that cosmos can do what takes a PhD student or a scientist six months to usually do with 80% accuracy. Kosmos uses a structured world model to share information between a data analysis agent and a literature search agent.

6. 1:12

   The world model enables cosmos to coherently pursue the specified objective. Over 200 agents rollouts collectively executing an average of 42,000 lines of code and reading 1500 papers per run. Our PhD is dead. Or is everyone soon going to be able to get a PhD? The world is changing fast.

7. 1:32

   Let's keep it open for everyone and private for each person. Our second story today, ByteDance unveils China's most affordable AI coding agent at just $1 and 30 a month in US dollars. The owner of TikTok continues to impress, this time with the most affordable coding agent in the world, and 62% cheaper than other agents in China.

8. 1:51

   Chinese AI developers, for larger context, have been rolling out new models at a rapid pace in recent months, including moonshot AI's Kimi K-2 reasoning variant and Mini Max's M2, which sparked comparisons to what everyone now knows as the deep sea moment that originally put China's AI sector on the map.

9. 2:10

   This is simply a continuation of the larger story we're tracking in that larger story. Closed source frontier models in the West are competing with open source models from the east, and soon, if NEAR continues to ship in terms of their AI development pipeline, NEAR will be complementing this landscape with neutrally verifiable models on NEAR.

10. 2:30

    What a time to be in the NEAR ecosystem, even if you don't know it yet. Last up today. Not all of the updates in AI are positive and exciting. The weaponization of AI is only the beginning. This attack, directly responded to by Anthropic, shows how hackers jailbroken and weaponized it to autonomously attack certain high value targets.

11. 2:51

    In the attack, the AI completed roughly 80 to 90% of the campaign autonomously, with human operators stepping in only for about 4 to 6. Key decision points. That means that we are very close to having fully autonomous cyber attacking AIs.

12. 3:05

    If these stories tell us anything, it's that the future is going to be unpredictable. And as a community, it's our job to remain educated and aware of both the possibilities and the perils. As we march towards Nearvana with a positive, human focused vision for an AI future, we'll do this all again real soon.