The big bet
I am referring to this concept only from today. I had this recurrent thoughts about how internet will look like in 2-3 years but it always felt the same. I think now we will more and more start to hear about web 4.0. Web 4.0 is the web of LLM, where not only user interact with users, but LLM can interact with other LLMs mostly but also other users (but as we will see later is a little counter productive).
How did it start
It all started from an iftar. Iftar is a tradition in Islamic culture, where people during Ramadan gather after the fasting period and shared the first meal after the whole day of not eating. Me and my team from our current startup Wellround were discussing some ideas about data collection and LLMs and suddently the serendipity happened: we were able to connect the dots. In the future, maybe already here, maybe in few months LLMs will probably talk to other LLM and get shit done. Not just coding, not just answering our questions, but performing actions on our behalf. At Wellround we are working on creating a digital twin of a person only from readily available data from smartwatchs, online services, daily reminders, calendars and what not. Turns out our vision was not big enough. We were told that we need to be able to tap into a bigger amount of data, there is where value is stored and not only, but mostly in the connection of this information, which is actually knowledge.
Turns out that we can do more than that. What will happen when we let 2 LLMs discuss between each other, having a different background? What happens when that background is the context of 2 different persons? Can we create some value between the interaction of two LLMs carrying informations of two different personas?
The most importan currency
I have talked in many previous blogs, how I find that the most precious currency and thing that every human being have is time. It is limited, period. We should make the most out of it, period. What if we could offload some tasks to LLMs? What are the tasks that we could offload to LLMs? What can we get out of it? More time exactly! But how do we make LLMs effective to do this tasks as if they were us? Well, they indeed need to know us very well. Our interests, our hobbies, routines life, work etc…
We build a second self!
Our goal at wellround has always been to be able to aggregate data from multiple parties with a big focus on health data. What we have figured out while working on this task is why cann’t we aggregate more data and what is the use that we can then do with it? With the increasing amount of data that you aggregate from a person, you can create a digital twin, what we call ‘persona’. The more data points, the more accurate the description, the better the results when they are given to the LLM. But what are the use cases? Well, just think of thinks you want to do to reach an outcome but don’t have time to?
- Planning a trip with friends? Offload it to the framework that queries all the personas that are involved in the trip so that they can accomodate everyone.
- Want to find a new partner but don’t want to doom swipe? Let the LLM do the heavy lifting and “chat” with potential candidate’s LLMs and get back at you with just a few “potential dates”
- Want to book a restaurant? Offload it to the LLM, and it will know your preferences, yor history, what you already had to eat that day and so on…
- Need a psychologist? prompt our system to llok for one, and the LLM will search and will “talk” to different ones and get back at you with the one that matches your needs and has an expertise on the things you might be needing.
How do we do that?
That’s the real question. The main problem is that all the data needs to be kept private and needs to be accessible by different applications in different ways, with different access rights. This is what we are currently figuring out. Another problem is that we need to understand where this data can be stored. Will it be a peer2peer network or will it be on the cloud with LLMs just querying this database on their own?
Stay tuned at Wellround.me to find out!
If you are motivate and brilliant SW, ML engineer and/or cloud architect reach out to me at sebastian.cavada.dev@gmail.com to create a real and valuable future together!