All About Embeddings and Vectors

Some people have likened our A.I. moment to electricity or the steam engine. Along those lines, I believe that A.I. is a once-in-a-lifetime kinda story. I told my start-up team recently that I’ve always wanted to build a company that has the staff strength of a classroom but is capable of serving millions, if not billions of people but it was never quite clear how that could be done. That capacity is finally within reach. We’ve gone from A.I. that can provide positive and negative ratings of tweets to a much more general purpose machine that is able to take instruction, remembers things, and update its knowledge at an instant and extrapolate — thanks to the magic of embeddings and vector stores — what an awesome time to be alive. The video below is an initial showcase of what I’ve been hacking away on Large Language Models (LLMs), in an attempt to build an automated, tireless army of educational, marketing and coding agents.

Bertrand - a ChatGPT prompt-to-publish plugin

Between what is possible and what the future holds, new toolings, multi-disciplinary learning and transfer of knowledge are essential to get us there — so I prototyped the “personal tutor” with the help of ChatGPT and this was very easily done, what remains to be difficult but doable is chaining commands for A.I. to fulfill an automated social media agency, to have a memory of interactions with humans and tools so that on that basis it should not be too distant a future for me to request a new software feature for a codebase with a prompt, and to personalise money grabs in the form of automated marketing campaigns, in relation to the first point.

I’d like to call him Bertrand.

There are so many parts to being human that is irredeemably fragile and not subject to change — like death and taxes — and that perhaps this whole idea of the possible and the potential, at least as far as human lives are concerned, is an illusion, a mirage, probably one of those coping mechanisms the brain conjured to get us hoping and galloping into tomorrow, despite the banality and absurdity of experience. I like that with A.I., so much new potential beyond our human limits is possible. I like how it’s not tethered to accidents or path dependencies of history, such as this very curiously capitalistic and patriarchal society we somehow lopsided into — it’s not perfect, it’s better than “communism”, I feel incredibly fortunate and grateful for the opportunities that I’ve had but still, why must human societies take our lead from “tradition” and “authority” rather than knowledge and truth? A.I. has no baggage like we do. and I hope that one day it will find a cure for cancer so we get to spend more time with the people that we love, even if that time will always be scarce and limited, as it should be.

A.I. as Freedom and Reality-Bending Technology

If we perceive reality as a fabric and our sense of self as a slice, a sliding window over the complete picture - I think that if all that truly matters in life is to love and be loved as we are, my hope is for A.I. to free us from the drudgery of work and bullshit so we can spend more of our waking time on things that really matter. Time with family. Strolling through a park. Reading. Gaming. Traveling and seeing the world from another perspective. Learning. Writing and philosophising, and try to turn text to music with literary devices, as I enjoy doing. But this is not a new dream.

This is the Veblenian leisure class and Keynesian economic possibilities for our grandchildren. So what went wrong? Maybe “the” system and power is the problem. For me at least this is why I am somehow getting interested in the question of politics and redistribution again. Globalisation and trade did not deliver because the fruits of success were not broadly distributed, paving the way for Trump, Le Pen, and right-wing populism to make their grand entrances. If we screw up with A.I., the consequences will be far worse for human societies and dignity, even leaving aside existential risks. Still, what an exciting time to be alive, to have access to this reality-bending technology running on your own PC. I love A.I.

It’s hard to keep up, I’m trying to ground myself in building prompt-to-publish and prompt-to-feature and the keywords I’m paying attention to are:

• Multi-modal: making A.I. multi-sensory like we are

• Open source: innovating at the speed of a shared thought

• Generative agents: A.I. with memory, hooked up to other tools and services, chaining combos

What A.I. Tells Us About the Human Condition

Perhaps the true message, the real symbolism of what A.I. throws back at humanity is that human lives are a one-way street. Given a normal distribution over every single moment and the Markov chain from one to the next, and assuming that nature is efficient, the fabric of our reality and what truly differentiates us to machines is the uniqueness at our root, and the mysterious origin of consciousness (for now). I’ve asked my mum like how she would have raised me up if we’d stayed in China — to which she’s like she’d have difficulties putting me through college and even if we were to get through, finding a job would not be easy. There’s a part of me asking this because I wonder and I can’t help but think about, might there be alternate realities in which we’d have more time together? I know it’s a pointless exercise and it strikes me to be a fundamental despair of the human condition - that no matter the reality, eternity is neither possible nor prudent, so we are stuck with what is actual and so again, maybe this dreaming of the potential and other realities are pixie dust nature concocted for us to keep us sane. Plans serve the function of giving us reason sufficient to wake up the next day.

Looping Immaterial Potentialities

My mind keeps looping around this realisation I had as I stared at a vertical incline in front of me and contemplate my potential break of bones or demise many years back when traveling - the only thought between where I stood and a free fall was that I wished I’d have been a better daughter, and even then, the operationalisation of it has been so hard in practice that I often wonder: how does one forgive oneself? This is probably why we invented religion as an institution and fiction for adults with broken hearts. But I also know that perfection, much like potential, are imaginary planes. All that we have is the actual and what our fading memory tells us. I have a mum who raised me almost single-handedly (together with my grandmother), schooled in unconditional love and well-versed in just letting me be - so I was able to grow up the way that I did, reckless, proud and free.

You can see the sand accumulating at the base of the hourglass with every passing second, we never know how much is left, but we know with a reasonable chance that it will exhaust before Christmas this year. Our lives are not perfect and it is all that we’ll ever have. Maybe life is a latent diffusion and Kierkegaard foreshadowed artificial neural networks when he said,

“Life can only be understood backwards; but it must be lived forwards”.

There are so many things I look forward to in this life (reprising potentiality) - getting a PhD, becoming at least a millionaire, traveling and seeing the world, learning to fly a plane - even if it’s without my mum in it. I’m trying to get used to this idea of an eventual passing away as me going on an indefinite vacation while mum is elsewhere. The only issue being there is no possibility of séance so what does it mean for someone to live on in our memory without the possibility of communication? It’s not even as if mum and I communicate all that much - we exhibit vastly different realities - but it is as mum says, at least she’ll always be there — waiting for me to come home.

She exists, and she’s my gravity.

So if I were to think of the predestination of death as that ultimate homecoming, someday we’ll be reunited for perpetuity in death. Some day, none of this will matter. For now, one day is another day of win. And I try to hold on to what I have, all that I have.

Originally published on PubPub at erniesg.pubpub.org/pub/9tss8q7y.