It’s been awhile! I feel like I haven’t had the time to sit down and gather my thoughts to continue posting. I’ve been so wrapped up in the forward motion of AI, and how it is impacting society more and more each day…

And then I get off twitter, and talk to normal people out in the wild (and by normal I mean, not chronically online like myself) and all they can do to describe AI is a “fancy chat, that talks to be about my day, and my feelings, and tells me things I google”.

This amazes me. There are millions (hell, BILLIONS) of people in the world who don’t see the rapid and exponential improvement AI (I hate calling it this. It’s an LLM. Its a really great NLP that has access to skills, and tools, and functions (which are all the same thing, but new buzzwords come up all the time these days)) has been moving in. I’ve tried to follow it all, I’ve tried to review mythos, and all the scary preview models that apparently will act like the best security engineers in the world. I’ve tried to keep up with the multi-agent orchestration, and the claudebots, and the agent loops. It’s hard. It’s fun.

But, I’ll never be as fast as a computer (Although, I’m not sure that’s true, I feel like humans have the neural ability to calculate more rapidly than a digital computer, but I could be wrong). And that’s ok.

Why?

Because a computer is a tool. LLM’s are tools. They are not replacements. They are EXTENSIONS. They are utilized to solve problems. They have a difficult time determining what problems need to be solved in the moment, in the future, or in the past even. They are constantly being updated and tuned, and controlled, and adjusted… over and over. At some point, we’ll figure out that the best way to have TRUE AI is to allow a singular JEPA to start in infancy. Then allow it to interact in the world (Like an RL system, similar to DeepMinds StarCraft model). Then give it strict rules. Allow it to break the rules, but punish it for immoral or unethical decisions. Repeat this. Do it in a condensed period that feels like decades to a model. Then repeat it again. Give it a different starting point. Give it less privilege, or more privilege. See what it develops.

The current LLMs (foundational models) are the culmination of all human contribution. They learned from us. They are trained on US. We cannot continue to utilize them to make predictions based on probability to create new things. They will always require NEW things to improve. They will always need US to properly utilize them to make the world a better place.

So, as scary as it feels…as crazy as it sounds to contribute to the what is likely going to be the largest job replacement technology in the world….

Use it. Build with it. Learn from it. Don’t trust it implicitly. Iterate. Learn from what it’s doing, and how it’s doing it. Build more. learn more. CREATE MORE. Allow it to become a tool in your toolbelt to get you to where you want to be.

At some point, the mega corporations will limit the access to these models to a select group. It’s inevitable. It costs money, and energy, and resources. It’s expensive. So learn to build it yourself. Learn to tune it yourself. Learn to incorporate it into society. If we all do a little, a lot is done by all.

Have fun. Dream. Play. Create. Don’t let my doom and gloom stop you from doing what you want to do.

But be vigilant. Understand the risks. Understand who holds the cards, and who’s thoughts/ideas/words/numbers/creations were used to train these models.

Yours. AI/LLMS/AGI will always be a combined ownership of humanity, no matter who tries to “meter it” or tokenize it. We can build the local models. We can run them on our own systems. Free. It’s ours. It’s built from us, by us, for us.

Until next time,

Haxk