• Just two years ago AI could only retain about 8 pages of information

• Now, it can memorize the equivalent of 10 King James Bibles

• Context window size determines how much models can remember, and it is growing at a staggering rate—1,250x each year

• Everything you do, see, or hear now fits as memories in an AI’s context window

• Combined with increasingly smarter models, this will be the ultimate competitive edge

• But there is also a risk of lock-in with platforms that monopolize your context

• Similar to how social media locks you in and prevents you from taking your friends and feed elsewhere

• An open protocol for portable context lets you move freely between AI apps without having to start over on memory

We started working on Palet with the mission to drive the adoption of open and decentralized technologies for contextualizing intelligence. The motivation to pursue this mission comes from a deep-seated concern for how the future of AI will turn out. We recognize that beyond the pursuit of smarter, faster, and cheaper models, the most significant differentiation will come from which providers can fully integrate your entire life's context into their platform. And as we’ve seen with social media, this always leads to an ecosystem where the winners dominate by locking you in and keeping you tethered. That is why we set out to develop an open protocol for building context-aware and personalized AI apps. Such a protocol guarantees that users can switch between apps while keeping their data across any service that utilizes it. And it also ensures that developers can build without being disadvantaged by monopolized context.

Among other things, we also aim to design a protocol with value streams that incentivize everyone to contribute resources. As it is the only way to ensure that we can maintain an open ecosystem that is also decentralized and durable.

Last winter, we started building our own client app along with the protocol. We haven’t yet settled on a name for the latter, but we’ve been calling the client Palet. It’s a browser that uses AI to capture everything you see, hear, and search for. And lets you easily retrieve information. We think the browser is the ideal starting point for building a great product around context, especially because so much of the information we generate and consume originates from surfing the web. Something can be said about our browsing habits too, and how they reflect personal beliefs. And perhaps how, as models get smarter, we can build personalized agents from it — incorporating your entire browsing context to form intelligence with similar beliefs. That’s the general direction we’re moving towards with Palet anyway.

But we also want to demonstrate that companies can build a business by offering services on this open protocol. Since context is stored on a separate, personal data repository synced across the network, apps that build intelligence on it benefit from each other. Meaning there are emergent, novel AI primitives waiting to be discovered. Ultimately, though, our vision of an open commons for contextual intelligence is not unique and is borrowed from ideas of the Semantic Web. The biggest difference is that the vision of the Semantic Web called for manually adding special tags to pages to make them readable by machine intelligence. By contrast, a Contextual Web can draw meaning and utility from data provided by the activity of the individual user. Since, as it turns out, AI (the machine) can understand things as we do. So there is no need for RDF, OWL, and other knowledge representations.

Anyway, we’ll be making our plans more transparent and sharing updates in the coming weeks. Not to mention, experimenting with different services to see what provides real value. If you’re interested in learning more or want to help out because you understand this problem space, feel free to reach out to us via Twitter, at @get_palet.