Topics: Technology, Business, Entrepreneurship
**John Coogan** (0:02)
Big, big news, Mark Zuckerberg dropped a blog post. Also, I mean, the bigger news is that they open sourced already one model, and they're planning to open source Muse Spark 1.2, that's big. But the blog post lays out his AI vision, and we should go through it. Meta was sort of set up to be, and also ran in the AI race, like the Llama project stalled out, Yon Lacoon exited. Their biggest model, Behemoth, never shipped, which was ironic because in the Bible, Behemoth is this untamable land creature, and of course, it was just the model was too big to be tamed to be released, and it was like nominative determinism in action. Happens a lot in tech, unfortunately. But they also turned over the entire team, and Zuck said, I'm not quitting. I'm not out of the game. I'm staying. I'm all in. Yeah. I'm raising debt, equity, doing all sorts of deals, making things happen, making big offers to AI researchers, rebuilding the team, bringing in Alex Wang, Fred Friedman, Daniel Gross, bunch of heavy hitters. And they already build massive data centers. They have a huge surface area to distribute AI, consumer experiences. If they come up with new ones, they have billions of users. They can put some, hey, try this thing, and maybe they'll get a bunch of users from it, if it's good. Semi-analysis also just one month ago laid out a much more detailed bull case for MSLs in particular. It was interesting because I was always optimistic about, okay, they have a lot of data from Instagram. They'll do a lot of images. They'll do a great image model. The image editing will be the best in class.
They can finally kill CapCut with their edits app, which is great.
When I think about you're taking an image, you're taking a video, you have a bunch of videos from a trip, you want to edit all that together for some Instagram story or Instagram reel. Reels editing is its own skill, its own talent. It's not available to everyone. A lot of family friends of mine will use Instagram, and they'll just post a picture on stories. They're not doing edited cinematic reels, but with AI, you can potentially get someone from, oh, you took a bunch of video, some of it's horizontal, some of it's vertical, some of it's sort of cropped wrong, some of it's colored wrong.
**Jordi Hays** (2:05)
Turn that into some brainwrap for you. You can just throw it right here in this trough.
**John Coogan** (2:10)
Yes.
Adding those programmatically. No, you actually could. You actually could. And so I was optimistic about that, and then obviously we talked to Eric Suford about just the gem model and how AI can immediately move the needle on ad monetization targeting all of that. So there were lots of reasons to be optimistic. Semi-Analysis laid out this more detailed bold case, and they said that there was this interesting detail that maybe was getting missed, which was that by recording the screens of every meta employee, which is very controversial and bold, but actually ripping that band-aid off and just saying, hey, we're an AI training company now, and everyone's going to be recording their screens and their flows and how all their work gets done. They basically got a frontier leading AI data labeling company for free. I mean, they took a PR hit for it, had to answer some questions, but in general, the volume of tasks, the volume of quality data that's coming from that workforce now is on the order of a scale, Mercure, Handshake, et cetera, Surge, and so they just get that more or less for free. Obviously, Meta's lawyers are great, Meta's researchers are great, Meta's finance team is great, so if you're getting all of that, it's equivalent to hiring a really talented person to do a training task.
**Jordi Hays** (3:30)
Yes, yes and no. I think a lot of the data that the Surges and Mercures are creating is much more specialized and I think there's just a lot of noise coming out of that data. They also have rolled some or maybe most of the program back. I think there was so much pushback.
**John Coogan** (3:46)
You can pause it for 15 seconds, once a day. I think you can pause it for half an hour, an hour, if you want to go get a nice scroll in at work. But I mean, that's on brand. If you're like, yeah, I was scrolling reels, I work at Meta.
**Jordi Hays** (4:00)
Testing, I was hunting for bugs, your honor.
**John Coogan** (4:04)
For sure, for sure. But yeah, I think there are some employees who are just fully excluded because they work on proprietary things, they don't want leaking into the training data whatsoever.
25 more minutes of transcript below
Thousands of transcripts fetched by people building searchable podcast archives
Try it now — copy, paste, done:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090
Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/YOUR_EPISODE_ID