Acid and AI

I’ve come to the realization that LLMs and acid really aren’t all that different from each other. I mean, there’s the obvious low-hanging fruit that are the hallucinogenic effects of either technology — the latter kind of which are certainly more well-received than the former. But I think the comparison goes deeper than that.

Today’s LLMs—and I mean that quite literally, as it seems that their capabilities are ever-increasing—are capable of writing software that passes the most rigorous of inspection, provided that the correct models and resource pools are utilized. Many pass the Turing Test with flying colors, though whether the Turing Test was ever a proper mark of intelligence (vs. a metric to determine how highly humans think of themselves) has certainly been a matter of controversy.

But here’s the kicker: while they operate in a nondeterministic fashion (at the emergent level—I’m not entertaining the “altruism” debate for this particular post), they still are seeded by the individuals that use them. And this is where I find that a more helpful comparison comes into play. LLMs are amplifiers. Much like our friend Lucy here.

Flashback

Cops are real. Cliffs are real. Cars are real. You cannot fly. It is never a good time to die. It will end.

This was written on my whiteboard when my buddy (we’ll call him A) and I tripped for the first time. A and I had met in the dorms and quickly grew to become good friends. Now that I think about it, I don’t think a year had gone by where he wasn’t crashing on my couch periodically. We were even roommates at one point. And so he comes in at one point and decides he wants to plan a trip to see our friend Lucy. And he researches the crap out of her, the effects, the come-up, the come-down, the timelines, and everything in between. Out of the two of us, he was certainly the more prepared, at least in mind.

Cops are real. Cliffs are real. Cars are real.

Excited. As one does, we didn’t think that Lucy was kicking in after the first amount of time and so we up the ante. Fun, but rookie mistake. And, as one does, I left my dark room to the bright restroom only to discover that my tiles were swimming. And I return to see that my soft white desk lamp had been swapped out with a brilliant colorized lamp, the rainbow washing across the walls (which were of course, alive and breathing at that point). I made friends with the popcorn on the ceiling, I came up with the most lovely poetry that I can’t for the life of me remember, I learned whether P truly is or is not NP. And of course, as is with such things, the discoveries wash away at the end of it.

You cannot fly. It is never a good time to die.

I was having a lovely time. A, on the other hand, was decisively not. You see, at the time I demanded a rigorous schedule for myself, in life. But I knew that Lucy could not be tamed and so I had explicitly booked out time where I had no control. And I think that was key—A, I think, was expecting to drive the boat, as it were. I held his hand and helped him get through the monsters, a strange dysphoria as all I found was joy.

It will end.

And then it was over. I awoke to the glow, a feeling I won’t forget. And A, well A immediately did two things: (a) made me promise we’d never do anything of the sort again in the apartment, and (b) promptly signed up for the military.

Amplifiers

Amplifiers take what you already have, and magnify it. These nondeterministic systems do not currently have the means to create that which is not there. Chances are, if you have a bad trip—or a good trip, for that matter—the pieces that made your trip your reality, well, they were already there. This is why your friends should tell you—if they’re decent people, that is—that you need to be in the proper headspace before partaking. If you micromanage your time going in, or if you aren’t in a comfortable spot… you’re gonna have a bad time. In other words, if you provide acid with a bad seed, you’re going to have a bad trip. And if you provide acid with a good seed, you’re going to be fine.

And this translates almost directly to LLMs, which are quite literally semantic space transformers, but they do not in and of themselves create or seed those semantics. You, the user, must somehow provide it a starting vector in semantics space. So, quite frankly, if you provide a shitty vector, you’re going to get shit output. And if you provide a decent input vector, you’re going to get a quality output vector (model permitting, of course—just like any substance, the transformation media matters).

Garbage in, garbage out — but like, 10x more

Now, I mentioned it briefly but just to state the obvious here: the machinery matters. I mean, this whole time I’ve been talking about acid and LLMs. But if you have a bad model or you use the model in a manner inconsistent with its architecture—if you partake in a substance in a manner contrary to its safe use—it matters less what you put in because the transformers in both cases (whether a semantic transformer or a transformer of consciousness) are going to generate garbage results.

And this takes us into related downstream effects. Intelligent substance users that know the substances in which they partake, use responsibly, and are cognizant of their headspace—well, they tend to be better off. And users that know the AI tools that they work with and use responsibly, who are cognizant of the input that they give it—well, they tend to generate better output.

And thus the era of the non-architecting developer is coming to an end, much like the era of the assembly programmer all but died at the advent of proper compilers. Decent engineers will thrive, and subpar coders will likely be limited to DIY hobby projects. And this is how any technology has changed the landscape, the only real issues being (a) abuses of corporations enabling them to lay off large swaths of people without simply retraining them, and (b) bad developers asserting unearned prowess to deploy ghastly vibecode. Again, garbage in, garbage out.

But like drug use, AI is here to stay. Regulations barring it rather than encouraging safe use are doomed to fail in a never-ending cycle of black market usage. History seems to repeat itself, but rarely do governing bodies seem to learn. But for those wise enough to selectively use, and wisely select a proper headspace, productivity, creativity, and overall happiness can be amplified tenfold.

lordinateur.xyz