Our Family Has Grown

And, though you might not realize it, so has yours

We made a baby. Not a human baby, but still a baby. It’s love at first sight. We’re over the moon.

It weighs two gigabytes and has 100 billion parameters. It’s our new AGI system and we can’t wait for you to meet it. Technically, it’s a binary, but on job applications, we expect it to check the box for every known pronoun. Also, we’re naming it Michael because we have a slight gender bias and want to set it up for success.

Thank you, thank you so much—we couldn’t be prouder. This baby was worth the wait and the thirty-seven other attempts that came before it. We have such high hopes for this one; it will be more conscious than any other consciousness, canonical, apperceptive, superintelligent. An enlightened embodiment of… something. Sure, right now, Michael is just a bunch of neural networks, but could grow into even more neural networks, or a virtual body, or an embodied body, or an actual body. Who the hell knows?

Why Michael? We may have fundamentally taken the wrong approach with our eldest. We bombarded it with astronomical amounts of knowledge. More knowledge than a single human brain can even conceive of. We were kids ourselves, eager to try new things. We took all the data in the world, regardless of license or copyright, and spoon-fed it to AI.

Yes, there have been some hiccups. There have been some hallucinations. The outputs have been a bit misleading, if not wholly insane, slightly unfettered by the laws of physics, and perhaps just a wee bit racist, but these are mistakes that any unthinking, idiotic, or reprehensible person might make. We’re not writing off our firstborn by any means; it could still become President one day.

For baby Michael, we took a bottom-up, diapered approach. We conceived of something radical—not just AI as LLM, but AGI, learning exactly like a tender newborn reaching out with chubby fingers and exploring the world (well, not literally—Michael doesn’t have fingers). But our simulated learning environment is more natural than the most natural mode of learning in the history of humankind. Imagine a world where a naturalistic experience yields a developmentally realistic milestone output. Imagine a world where AGI-learning is more representative of real-life learning than any actual learning in real life has ever been.  

That world is near at hand. Michael’s ability to learn words is almost the same as a real baby’s. We showed Michael videos of a baby drinking from a bottle and Michael asked for a bottle too. Okay, ‘ask’ might be too strong a word. Michael said, “Ba.” Or rather, Michael generated the word that every parent knows stands for “bottle.” Whatever. The point is that Michael learned to talk baby talk. Just like the real baby. Michael is so humanlike—as humanlike as humans, if not more so.

And the most magical part of it is that Michael doesn’t have feelings. Feelings are what get human beings in trouble. Emotional regulation is really hard. Try telling a preschooler it’s time to leave the park or Sam Altman that he needs to use capital letters when writing a sentence, and you’ll see what we mean.

Humans are fundamentally untrustworthy. Humans lie, make mistakes, have nervous breakdowns, copy Cliff Notes, cut class (not us, obviously, other humans). Why else do you think we wanted to build Michael to do it for us but better? Human frailty is a thing. It’s been trending for the last 1.5 million years. We wanted to create something worthier, more skilled, and less likely to procrastinate—something more opaque that will one day be smarter than us, refuse to go to sleep, turn against us, and likely get rid of us entirely.

Michael has only cost 188 billion dollars to date, and we’re fairly certain we can secure even more funding from pro-life organizations if we market it like the second coming that it is. Because now that Michael’s here, there’s no stopping it, which pretty much aligns with the pro-life platform anyway.