I was just talking to my friend about basiux, per his request, and he suggested me again, like everyone I talk to about it, to get in touch with you guys. Of course, I said "already did, and they put me down":
After reading your email again, Matt, I must say...
What do you even mean by accomplished?
And did you notice "my mario video" you talked about wasn't mine?
I'm improving the way I communicate by the day, and shouting out whatever I'm trying to say while representing basiux is really hard. I'm not sure how long you guys took to write that landing page on openAi, it is very well written but... It's also still too subjective as well. Leaving way too much of a margin for interpretation and a couple of questions...
So, let me put this in another way: I'm not sure I even want to join you anymore.
Why aren't you going full open about it? Wasn't it supposed to be open? Well, you're around for at about 1 month now and you're probably together for longer than that and you released only 2 static pages up to now.
Would you say we need to be able to control the Ai in any format? I personally think that's the most dangerous aspect we could strive for in an Ai and the letter is far from clear in this aspect. ASi must be set free and we must fear nothing from it or else, that's where the only danger may lie. Although, the most dangerous aspect of it is losing precious time.
Now, I'm pretty positive that with all your team's technical background and accomplishments you'll still have some hard time understanding what I'm trying to say here. Mostly because, well, I'm terrible at making myself clear. I have always been.
Since you're swamped and open for questions, I just did think in a very important question we'll face and you probably are light years ahead of me and maybe every other group out there:
What would be your engineering approach?
Unfortunately, I'm not versed into the literature. I haven't even had the chance to dig down pieces of code myself, but... Our current approach would also be along these lines:
Some kind of genetic algorithm, reinforced by just a few different fitnesses.
Many different graph sets / neural networks to analyze input from different angles (analogous to emotions).
As stated elsewhere, the subconscious would need to be constantly compressing data (i.e. recognizing patterns)
I was very glad just now, after reading the first link there. We are indeed completely aligned and I kinda envy your ability with words, as a team. Very well placed answers, great interview!
It's really tough bringing these ideas to so many questions that are so unaware of what Ai even is today. Just by doing that alone you're kinda already helping us as well. We can always point to your research!
basiux.org has the same goal as openai.com with 2 twists: I started it when I was born. And I've been using all my free time since July into building what you can see on the website now (along with all links available).
All that is being translated into the novel. He told me he was reading some of my blog or something and he thought to himself "fuck, this guy is writing the next Matrix!". Am I? I sure like to think so.