WOYM: Meta Llama 3, Productivity Boom, AI & Datacenters

Guests:
Ram Ahluwalia & Justin Guilder
Date:
04/22/24

Thank you for Listening this Episode!

Support our podcast by spreading the word to new listeners. We deeply appreciate your support!

Episode Description

In this episode of What's on your mind - we discuss Meta's newest upgrade to Llama, Datacenters, Energy and AI wave.

Episode Transcript

All right. Happy Monday, Justin. 

Justin: Hey, Ram. Happy Monday. How are you today? How was your weekend? 

Ram: I had an Indian buffet in New Jersey. I'm like still stuffed. Since leaving New York, it's hard to find, or at least I thought it was hard to find good Indian food. Apparently, not the case. It makes sense.

There's a bunch of Indians in New Jersey. And so a friend told my wife at the spot we went there, it was a, All you can eat buffet. It was, I had four servings. This Indian dish called Ki, it's like a rice pudding. I actually took more to go and [00:01:00] was slurping it as if I was like on an IV drip while driving home and I had a massive food.

I haven't done that a lot. I had a massive food coma. My cheeks feel a little puffy today. I definitely gained some, a little 

Justin: extra salt. 

Ram: A little what? 

Justin: A little extra salt. The salt sometimes. And like too much food gives 

Ram: you a little extra . You got, I think I've seen this hat before, by the way.

Justin: Yeah, that's true. I, I do have a new hat in the closet, but there's a lot of hecticness at my house right now. I have a broken AC that's getting fixed. I've got painters who are in the house trying to get my house set for moving. I've got one kid homesick. One kid on a planned school day off. So I pulled up in my office with my dogs and decided that I would do a repeat rather than exit the office.

Ram: That's a, time for, when I see two hats, then I ask myself, is there an intervention required or noT? 

Justin: Next week we'll have a new 

Ram: one. Okay, good. Let's get into it. A lot happened. In the last week, even this morning [00:02:00] breaking news from Meta, they just dropped the Horizon OS operating system.

You haven't heard of Meta launch an operating system. So I'm going to share my screen here. This is an excerpt from the Lumida ledger. I'm proud to call it one or two elements from the ledger. Which goes out every Sunday, but, let's see if you can see it on the screen here. Meta, which is one of our, core positions, they launched LLAMA 3.

It's as strong as GPT 4. That's a big deal. It's really good. I was playing around with it. This weekend, by the way, it integrates with Bing and Google search, and it feels more native, meaning when I'm using GPT, it'll say like accessing Bing, and then there's a delay. When I use Gemini, it, for some reason, can't even access Google search yet.

I'm sure they'll fix that, but Meta does both, and I can pass it a URL. I can say, summarize this, which is actually a killer app in [00:03:00] AI. I'm very impressed. These guys have distribution. And they've integrated it and are integrating it in WhatsApp and in Messenger. We'll see what they do with Instagram.

So unlike GPT 4, where they got to sign up new customers, they have the distribution and they're charging all those 0 for this. They spent billions of dollars investing in this AI and it's for free. For consumers and you and I can go use it and build business applications on top of it. This is a pretty decisive moment in the history of AI.

Justin: Classic example of kind of reach customer reach verse innovation. And here you have somebody that's at the forefront of both, which is dangerous. Normally you have incumbents that have. Reach and new entrants that are innovating faster. And here you have a incumbent with [00:04:00] massive reach that's innovating, perhaps not as quickly, but quickly enough.

Ram: I would say faster. Look, they cut expenses in 2022. Then per Zuck's interview with Dorkesh a few days ago, they didn't exactly see AI unfolding in this way, but they took decisive action. bought a bunch of 8100s from NVIDIA. They hired, of course, Yann LeCun, who beat the Turing test. They've got incredible engineering and it's fast.

So it's impressive what they've done. OpenAI is caught in the middle. They, the firms with more reach and better go to market usually beat the firms with the better product, but Meta's got both. Right now, Beta is in the pole position. Guess what? They're training 400 billion parameter model.

It's training right now. It's not available yet. It's already scoring better than all the current models. There's this, eval score. They're a bunch of evals, but one, uh, think of it [00:05:00] like a zero to a hundred. So the best models today are like in the low to mid eighties on the score. We define the score in our Lumina ledger.

This model is clocking at 85, which is the best, and it's still not optimal. It's still getting better and Meta has data, all these videos people upload, it's not as broad as a YouTube content library, but Meta does have data, Meta has all the posts, Meta probably understands like human nature and psychology better than most people, most other types of LLMs that we built.

Justin: Yeah, from a human psychology standpoint, Meta is sitting on top of incredible data with Facebook and Instagram. What keeps people engaged? What type of content do they read? So they absolutely understand that. That's one of the scarier parts about what AI has been training on and how the influence of people can be easily aimed.

Ram: Therapist's going to go out of business. Therapist AI [00:06:00] is coming. 

Justin: That might be a good thing for humanity to have some more therapy and reduce some of the anger. That's 

Ram: a great answer. It's a great thing. It'd be great if the therapist AI can talk to the other therapist AI that's dealing with whoever someone's having friction with.

That'd be the best. Like a couple of counseling therapy AI, right? Oh, there we go. So we have these video outtakes, by the way, on Lumida Wealth, these, the highlights of Zuck, being interviewed, so if folks want to take a look, please do. And then here you can see, the meta AI in action. Probably about meta.

Grand master plan. So I believe Meta's grand master plan is actually to, to really, go after the operating system, and, here we go. Okay. I don't know if it really told us the grand master plan. Probably not. They just gave us the vision, but, so they launched horizon OS this morning, which powers the quest.

[00:07:00] Virtual reality app device, virtual reality device, but here's the thing, right? Two or three years from now, if not sooner, these AIs can spin up their own virtual dashboards. Right now the interface is text, almost like a chat screen going back and forth, but there's no reason that the AI can't spin up a visual interface.

In fact, it already spins up visuals today. We call it like Soma and GPT or like Dolly. You get an image, but it's not interactive. You can make it interactive and there's some basic interaction. We'll say, Hey, pick which of these four images you like the best, and it'll make that high res. So if the A. I.

can spin up a spreadsheet or a calculator, why do you need or a document or processor interface? Why do you need Microsoft Office? Why do you need all these other apps we have now you'll see the collaboration apps like slack for some time, right? Some are easier to displace than others, but it's not clear that, Microsoft, which is the consensus AI name Which is like a 95th percentile [00:08:00] valuation.

I don't know how that changed over the last week or so. It's obviously Taking a bit of a hit. It's reporting earnings this week. I think Microsoft actually has the most to lose because they don't have data in the same way that Google has or Meta has and data is the moat. And no, number two, Google and Meta have that consumer distribution reach like we talked about earlieR.

 Yeah, I think that's actually the greater risk. It's not the death of search or the death of Google. That was all overplayed. Google's near its highs now. So anyway, that I believe is what to watch. We have to see when Microsoft reports this week, how is Copilot doing? Are they getting enough funding?

Traction, because the stock is pricing in a lot of earnings growth from co pilot. 

Justin: Yes, I'm really interested to see that because I'm still trying to understand whether these types of AI systems will operate parallel to one another. Like I still would use a Google [00:09:00] Workspace AI to optimize things in parallel.

Docs, slides, sheets, and in a personal life, something else. I'm off social media these days, so it probably wouldn't be, the meta version, but I can imagine a world in which I'm comfortable toggling between multiple AIs. That have better capabilities in personal versus business. And you can think about that, just like you said, from a data perspective, some of the data that Facebook has may be much better at helping somebody navigate personal item versus somebody navigating a business question.

If I have every Outlook email that an organization has ever sent and every business has ever sent, I might be able to help you navigate personal items. A conversation with, a colleague. On a tense budgeting issue, verse, how do I talk to my, college kid and you, meta [00:10:00] AI, know what every college kid that age wants and is telling their friends, and I can be a better parent.

Ram: Yeah, look, I think we'll have highly specialized AIs. Consumer will be one vertical, then of course enterprise and government, then within those we'll have different, but at the consumer level, we'll all have a primary AI and whoever owns that primary AI is going to get the lion's share of the reward and they're all gunning for it.

Here's another comment Zuck made. He said, our goal is to build the leading Open source AI. So leading means the best they want to beat GPT. That's what, that's one of those things. What's in your hair. It's like obvious, of course, that's a strategy, but what people don't treat as obvious is that open eyes, pull position is secure.

That's I'd argue that's at risk. So leading it is right now. It is the best. You can debate the benchmark. It's at least it, by my estimation, it's the best. We consider UX, not just the eval scores, even the evals. It's. It's excellent, the [00:11:00] integration across the app. That's their goal.

And, I believe Meta is going to build that hardware ecosystem, just like Apple, they got the Ray Ban glasses coming. So we have omnipresent AI. Now, then what's the next device, maybe a phone. They probably have to build on Android. By the way, Apple has been denying Meta the ability to upgrade certain features in the App Store because they have that authority to do that.

Again, it's anti competitive behavior,