House of Stake Livestream, J1 Teams Ready, OpenAI Using AI Code, Stanford AI Alignment Warning

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

1. 0:00

   Good morning Legion. It's a busy Friday in the near ecosystem and there's quite a bit to cover, so let's get right into it. First up, the House of State Governance live stream is today at 2 p.m. UTC time. Listen to the candidates, ask questions and decide for yourself which of the three is best suited for the role of Head of Governance.

2. 0:19

   Second, yesterday Illia dropped to vote in the Legion Telegram pertaining to the lock period for House of state tokens with 45 days taking a substantial lead. Third, there are now over 90 teams fully funded and ready to race J1.

3. 0:35

   The prizes include 5000 yea millions of Jambo over the 21 race season, and there are rumors of betting markets becoming available for different teams as well. We definitely hope to see you race in this ecosystem wide event. Last but not least, we have the missions to reach ascendant dropping on Monday alongside the Legion Jet book.

4. 0:54

   This is the first full SBT launch where successful completion of these missions will help you rank up for the first time. Let's get into two prevalent eye stories on the day. First up, The original AI 2027 report that we spoke about a couple of weeks ago had a 2026 forecast included that had some of the following predictions.

5. 1:16

   Coding automation goes mainstream, agents work like teammates, and AI. R&D is 50% faster for algorithms, not just compute. Well, it's still 2025. And very recently, an open AI employee say the following about their agent builder. Quote.

6. 1:33

   The agent builder we released today was built end to end in under six weeks, with Codex writing 80% of the press. End quote. Is it safe to say that coding automation is going mainstream? I think it is. And frankly, we're ahead of schedule, folks.

7. 1:48

   Second, and last up for the day is a study straight out of Stanford. It talks about how AI performs differently based on its specific context and tasks. When LMS compete for social media likes, they tend to make things up when they compete for votes in the political sense.

8. 2:05

   They stir dissent and they fight. When optimized for audiences, they become misaligned. Why might this be the case? LMS are trained, in the words of Brian Rommel, on the sewage of Reddit and Wikipedia. Is this sustainable, or are there potentially better solutions to training models and monetizing them somewhere in the future?

9. 2:26

   Perhaps this is another use case for years de cml. That's all for today, folks. We'll do this all again real soon.