Meta.ai has landed. The first draft. Straight stream of consciousness April 18th 2024
Story by Rob Tyrie
Warning this is a breaking news story, and this is my direct detection and creation of prompts to create a number of stories from this that are more possible. That means it’s full of typos and there’s probably some problems in the auto correction as I dictated it. I got to publish this now and keep updating it to keep it more polished but I wanted to record it on April 18th when all this happened…
Today social networks changed forever because AI has been released to the population that uses them in an unconstrained regulated manner. I think this is a bigger experiment than anyone has ever planned because it’s not really been planned. This is happening in an evolutionary basis. This attempt to connect a billion people through an abstraction layer that intercedes and interacts alongside people will have unexpected outcomes . And those outcomes will have the good The Bad and The Ugly over the next 1,000 days.
We’re in a unique spot in the proliferation of this technology and capability in the world going up an exponential curve. And in this case like all good systems that work based on metcalf’s law they get more and more powerful the more nodes there are the more edges are the network. This is definitely an AI Network. Is not any application anymore. And it’s a network that was designed to addict humans in order to maintain their attention in order to present advertising to them that they have been proven to be influenced by. The market hasn’t lied. Because it never lies.
So what happened? Meta. Com released metadat AI across most of its social channels and as a plan to release it across all their social channels to all their users in every way shape and form they use meta properties. I’m wondering if Mark Zuckerberg think this is the way to build the surface or the initial layers of the metaverse.. it’s an AI universe and it has been so for a few decades. It’s not reaching the edges. I’m sure lots of questions will be asked and now the policy people and regulatory bodies have to really figure out what to do now.
And maybe it’s nothing maybe this is just a more efficient way for Facebook to compete with Google for advertising dollars as they fight for the front of the market. It’s definitely a different race. But let’s drill in to the things that are important that happened in the impact of LLma being released in this fashion.
The amazing thing happened on April 18th. Yesterday meta released metadat.ai across this platform channels such as WhatsApp Facebook and Instagram.. and it was based on Lima too today they rolled out Llama3 earlier than expected. It is very smart and comparable in many ways to the leading Edge paid for foundation models from OpenAI, anthropic and Google.
Initially people will be doing their multimodal tests and just comparing it as if it was another version of open AI ..
But it looks like a very much different thing in Facebook world.. I mean meta world. Billions of people use those platform channels and now ai is in a bit integrated as a content creation content generation content manipulation and confident transformation capability that’s really easy to use. It just looks like a message and you can add and generate things and change your words as you need them.. and all I’m on the way generate a meme that makes your friends laugh… because it’s hyper General it’s because it’s hyper-specific and in the control of the group. The influencers are going to have a heyday with this.
So this is a triple win.. A great Foundation model … that has a utility to it comes from a stable lab that has plans to keep going further and further. The model is a special kind of mixer expert It is multimodal and does text to speech and text to visuals it even does animation. It’s got a lot of business plans being rewritten right now. Small competitors will have to follow not lead here. And based on LeCuns vision and leadership yeah llms of the future will have some kind of world model built into them. Remember that these AI tools to build the large language models and the large mixture of experts that are multimodal are very hungry for data. The first they consumed is what all that was available on the internet and they’re still hungry. One of the things that is very true about meta is a content generation machine. Every conversation every share everything that is written on it everything that is sold on it is recorded.. and people take a lot of photos on Facebook and Instagram to share ... The newest most advanced short message sharing platform and threads and they didn’t even roll out on Threads yet. That anti Twitter crowd is going to gobble up Ai and use it for their purposes to amplify themselves hopefully in a positive way like it’s been going so far. Remember there’s no advertising on threads yet the experiment is occurring now. Influencers will have ai at their fingertips and disposals from massive brands like beastman, or whatever he’s called, and Beyonce and Taylor Swift, down to my friend Brandon he manages properties up north who’s going to use this to generate ads and he’s not going to buy software to do it he’s just going to use Facebook and AI to influence his prospects and delight his customers. That’s a lot of data and it’s a very big mess for the AI index to not include it in its near-term futures. The lack of data for llms is a past perspective not a future one. They just have to be attached closer to where data is created and a lot more data is created on Facebook than it is at Harvard University. Note that this is the first AI company that’s done the non open source open source so they control the end results whatever and they watch what everyone uses it as a deploy it, but they control the secrets of it as they keep scaling… it’s moat with bridges that control. The are definitely in control. You can tell by reading the meta.ai TOS. Read it carefully .
TWO. It’s rolled out with an end user consumer point of view it is simple easy usable and embedded in social media some of the most popular software the world has ever seen and certainly the most impactful and for meta a literal gold mine if not plan to mine about advertising Revenue. This is the shocking thing, at scale the first trillion dollar company has rolled up advertising combined without artificial intelligence wrapped up in conversations that are highly personal and highly surveilled.
This is peak surveillance capitalism probably and is a huge moat.
What I alluded to in one and two is very very true this is number three.
Three — data sets right now with our current channels they have real time data that is not only facts and a bunch of fake news and falsehoods but it’s continuous conversation amongst people and groups with companies brands and governments. And I’m talking about all the interconnections across the channels when you think about them connected Instagram, Threads, Facebook, WhatsApp, and more coming. With the philosophy of the AI Lab at meta, they want to connect to more data sources and connect to more sensors.. that’s like the Ray-Ban glasses and the Oculus devices. It’s not as good as robots but they’re darn close. They must have a strategy of course to use not only this data for point-to-point but all the attribution that goes along with the metadata, the tagging the likes, the Emojis the stickers along with the social graphs they really see the world the special place. This all gets amplified with a mix of synthetic data and synthetic tagging and true tagging as they build sets with the forethought of building the Next Generation large language models. There will be a race for content and content deals to feed this with New creations and the reworking of all creations to create larger attributed data sets now that we know how they work better. And now we know how to create separate experts and buying them back together. That depends more data. When I add multimodal needs into that that go across the two senses that have been covered three if you can include language.. depending how you counted there are other five or 25 more senses to add to these models to replicate or imitate how human brains are educated and trained and how they survive in the environments that they’re dropped into. That’s a lot of data too. What’s different at this stage is they know how the data has to be shaved to improve the results of the model I just memerization or just good enough generalization. This is will lead to the current ideas and capabilities of all large language models. They’ll have to come up with the new name because it won’t just be language as Google has already proven with it s video processing
That’s three things. Right now if I think about Google he doesn’t have the advantages in the users with the social graph the Facebook house. Facebook doesn’t care about search it cares about social interaction and the tension people have to their social networks.
There’s probably a fourth but I have to think about that a bit and draw a diagram.
My next experiment is to do five things:
- Post about this and identify that Llama 3 is available to my community
- Connect with smart people and talk about this because it’s important.
- Read and post about this using perplexity and gamma.com
- Generate a chatbot a custom GPT and some surfaces to process and transform this news.
- Practice with llama to see how good it is and compared to Google open AI and
