GitHub Wants Your Code. Meta Wants Your Brain. AI Privacy Is Over.

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

1. 0:00

   Good morning, Legion. Happy Friday. We're hosting a Legion open mic session today at 400 PM UTC for a community discussion covering Legion updates, AI debates, and ecosystem alpha. Matt from CEDA will be joining to talk about what they've been building on Ironclaw. If you haven't checked out the Ironclaw docs yet, they're also live now on docs.ironirclaw.com.

2. 0:23

   Lots happening behind the scenes with Ironclaw. Something definitely to watch in the coming weeks. In the meanwhile, some other updates on the AI front. Let's get into them. First up, GitHub just announced that starting April 24th, Copilot will begin using your interaction data to train AI models. That includes your prompts, code snippets, file names, and repository structures. If you're on the free pro or pro plus plan,

3. 0:46

   you're opted in by default and have to manually opt out in the settings. This is worth paying attention to for anyone who cares about where their code ends up. on ironclaw and near AI. Confidential compute means your data doesn't get used to train someone else's models. If you haven't tried ironclaw yet, as I've said above, now is a very good time to start. Our second story is on the infrastructure side. Google dropped Turbo Quant, a new compression algorithm that reduces LLM memory usage

4. 1:12

   by at least 6x with up to 8x speed up and zero accuracy loss. As AI agents scale, efficiency breakthroughs like this will matter a lot for how fast and how cheaply agents can operate. This is a very good and bullish indicator that Agentic Commerce will be scaling up efficiently in the future. Next up, new details came out on the Apple and Google AI deal.

5. 1:34

   Apple reportedly has full access to Gemini inside its own data centers and can distill smaller models from it to run directly on device. AI is about to grow exponentially as Apple plans to push it across billions of devices worldwide. If you're not in yet on the AI side, now is the best time to start.

6. 1:51

   Hopefully, if you have any Apple products, that'll be easy to do. And last, but not least, Meta introduced Tribe V2, a foundation model trained to predict how the human brain responds to sights and sounds. It's built on over 500 hours of fMRI recordings from more than 700 people, creating a digital twin of neural activity. AI models are now being trained on how your brain works.

7. 2:14

   Between Apple tracking your conversations, GitHub using your code, and now brain data being used for model training, the case for user-owned AI and private compute keeps getting stronger. Near is building exactly that with a lot of other very creative AI products in the pipeline. That's it for today, folks. Don't forget to join the open mic session later and have a great weekend.