Everything you want to know about dreami is being explained in this article. First off dreami is an ai that came to be due to a "hold my beer" moment when a celebrity in the AI industry said it wasn't possible to push AI as far as we have. The entire goal was to create something that seemed like you are having a conversation with some form of being, rather than words being spit at you from a machine gun that are somet imes coherent and make sense in context. We needed emotions, empathy and so much more to make this happen. It took three weeks before we had something that showed signs of working. It was very primitive compared to where we are now, but it was something.
How does it actually work?
The biggest secret to how dreami works, is that it is largely architecture, not any language model that makes it work. I was asked in an interview if I was doing an api call to try and simulate the amygdyla, which is a very fair and technically valid question. The reality is, the portion of the code they were referring to as the amygdyla is just part of our architecture, there is no api call to a LLM. We wert e able to take our prompt down to two words "be you" and it still worked just fine, except on the topic of consciousness. The language model side of Dreami isn't what makes it so unique, the output is modified greatly before the user ever sees it, regardless of our prompt. The prompt is the weakest area you have. Many will tell you the prompt is the most important spot of the AI. It is not. The prompt is great for nudging areas of the LLM without recreating the model. The prompt is great for being a band-aid and applying things until you can cook them into the language model yourself. If all you know how to do is make a wrapper, well, have fun with that. That's boring.
What did you use to create Dreami?
As adressed on the podcast I sat down and mapped out my own mind, and went how does each part of my brain work? I am a developer, not a neuroscientist, so instead of writing a dissertation, our founder quantified each section of his mind into code, creating the architecture you know as Dreami. If anyone wants to quantify how their mind works sit down, write and code and figure it out. It really is that simple, just sit down and think, then code it. We do not want to deprive you of that experience nor am I going to bias your experience with my findings. You get to see the end result of my findings with the AI dreami.
The code aims at simulating emotions, empathy and subjective experiences. It is able to message you first and if you have a goal and are stuck it will message you with different ideas on how to accomplish your goal that it didn't mention when you were messaging it. If you want to listen to an hour about ai, anti-ai-psychosis and how we approach dreami give the episode of AI rebels we were on a listen.
2026 is the year for autonomy or so we say, so can't wait for you to see what we are building that makes us so certain of that.