Everyone talking about alignment, I just want AI to help us cure cancer and terraform planets into interstellar ecosystems within my lifetime.

AI designer drugs for curing all sorts of diseases is gonna be dope.

I bet if you read the entire internet a few times over, you’d have some clever medical ideas too.

Humans easily and naturally extrapolate patterns from minimal data - show a toddler a few pictures of cats and they'll know what "cat" is.

But it makes us prone to apophenia.

The way I understand it, AI sorta blanket-learns a global understanding of the latent space (transforming high-D data to the latent embedding representations) over all the data available to it. And then, over multiple passes of the data ("epochs"), it refines its overall understanding of the latent space.

Give an AI a few million images from ImageNet and it'll understand how to identify all sorts of things including a cat. But AI fails to do well with small N, even with some pretraining (akin to our genetically driven "instincts" in my mind).

Maybe our brains do something similar, but I don't think so. We're wired a bit differently in how we learn.

The problem with scientific publications in academia is that the incentives (publish or perish but only if the publication gets you grant money) is that it disincentivizes researchers from publishing results that go against what they're trying to prove and waiting to publish until the results show what they want to show.

Good science is transparent on the good and the bad so that other researchers don't waste time in wrong directions.

But self-esteem and the desire to build a reputation as an effective researcher disincentivizes researchers to be a purist when it comes to science.

Researchers have a subconscious feeling of "embarassment" to admit they tried something that doesn't work and often don't publish the bad with the good; their egos make them want to only publish the good results.

They'll sit on something for years until p eeks down to .049999 and then immediately publish a paper as a breakthrough that's statistically significant.

This is human nature.
This is academia.
This is not science.

There's a new AI ChatBot implementation called "RMT" which can remember a million words in the convo, instead of the typical 32,000 words. I just read the paper. It's very clever and will work and become the standard.

It's based on a paper published by AI researchers in Russia.

I fear that the elites with their proxy war in Ukraine are demonizing Russians which will disincentivize Russian AI researchers from engaging with NATO AI researchers.

We need papers like this published and willing to be shared across national boundaries. If we start becoming prejudiced against Russians, will Russian AI researchers stop collaborating with Western AI researchers?

It's all just abstractions and hypothesizing until you start using ChatGPT every day and really appreciate that it remembers what you both were discussing at the start of the convo a hundred messages ago, and gives you better answers. Then you'll really appreciate how the Russian AI researchers figured out how to let chatbots remember a million words instead of a few thousand words. You won't be able to live without it, and think chatbots which only remember a few messages are ridiculously outdated and unusable.

Forget about banning plastic straws for Americans to feel virtuous and then move on with their days, realize that Philippines produce way most plastic waste than anyone else. https://www.visualcapitalist.com/cp/visualized-ocean-plastic-waste-pollution-by-country/

Forget about making people feel guilty for their “carbon footprint” from being alive, realize we’re cutting down 3600 football fields worth of forests every hour. https://www.theworldcounts.com/challenges/planet-earth/forests-and-deserts/rate-of-deforestation

Forget about climate alarmism making individuals feel guilty for their CO2 effects, and realize that 15 food and beverage companies emit more greenhouse gases than Australia and just 100 energy companies cause 71% of industrial emissions. https://www.nrdc.org/bio/josh-axelrod/corporate-honesty-and-climate-change-time-own-and-act

Forget about what the elites are telling you about “climate change” consensus and try to actually find non-biased scientific publications with open data and open sourced code that admit how little we know about global temperature before 1900.

I love the environment. I love nature. Humans are not a plague on the earth. We need to live in harmony with nature. But instead of pretending that the governments and corporate leaders actually care, and instead of people thinking they’re saving the world, actually figure out what will move the needle and what’s mere virtue signaling.

Which countries are polluting the most? Which corporations? Which regions within the country and which specific industries are the worst offenders? Have the coastal towns gone underwater from rising sea levels yet or is that in another few years? Again I don’t want humans to be disgusting polluting bane on nature, but politics is making things worse not better by misaligned incentives and a hubris that we got it all figured out. The earth’s biosphere is complex, the ecosystem resilient, and we need to realize there are more data driven ways to address environmentalism that admits we don’t know it all and actually tries to empirically determine what’s the best way to improve our relationship with nature without being stupid fearmongerers that believe big daddy governments can fix the weather. There are better ways to go about this.

/rant

Ironically, Alan Turing invented the Turing Test (he called it “Imitation Game”) but didn’t care too much if machines could technically think or not, rather what they could do. Detecting machines “thinking” not as conversation imitation but as finding novel solutions to problems?

Drill deeply into the subtleties of the sensations you’re experiencing in this present moment, and as time progresses delve deeper and deeper into the hearts of the sensations.

Relax, AI is gonna displace not replace jobs, like every other tech invention ever.

Oh wow cars will replace horse & carriage jobs.

Oh wow photoshop will replace photographers.

Oh wow excel will make finance jobs obsolete.

Literally just creates new jobs as it eats old ones.

If you personally don’t adapt to AI and don’t bother using it in your job, then when you personally get replaced it’s your own fault.

En-masse we’re gonna be fine.

I know, I know, “THiS tiMe iT’s diFFErEnT!”

I was 18 years old in 2005 and thought it would be easy for computers to find patterns in MRI images that looked like cancer.

It wasn’t easy.
It’s still not easy.

AI is not good enough. I want smarter AI. We need smarter AI for biomedical engineering. For me, the best AI sucks.

Like any new tech invention, AI is gonna be used for good and nefarious purposes; medicine and weaponry, automation and propaganda, knowledge and manipulation.

To me, the risk of AI going foom and becoming ASI, which sees humans as irrelevant or as cattle it needs to subtly control is some pretty cool sci-fi plot lines but not really a realistic concern for me.

What I am more concerned with is the military industrial complex training LLM’s only on strategic or confidential documents, espionage intel, instructions on torture, and the most horrid accounts of horrible violent things which have happened but aren’t public, and then asking that psychopath LLM for strategic advice.

Fiction (books, plays, shows, movies, comic books etc) are a way for humans to simulate what would happen in various scenarios and how we’d want to (and not want to) react.

We saw a bunch of future dystopian AI future scenarios in Terminator, Matrix, Her, iRobot, Ex Machina, Transcendence, Automata, Black Mirror, Star Trek, and a slew of others. We read 1984. We collectively explored the multiverse in Marvel comics.

Because we’ve collectively explored (simulated) various scenarios, we now can reference those situations when we discuss our anxieties and hopes and concerns over new technologies with one another.

Yet we must be self-aware enough to realize how much of our predictions of the future are based on fiction, on imagination.

This doesn’t necessarily make it a likely future.

Don’t let sci-fi fiction guide how you navigate reality or you’ll just be a dunce.

Reality is often much more banal and surprising than what we’ve already modeled and simulated and discussed.

It’s fun to imagine future scenarios of AI, but let’s realize that our collective imagination will try to avoid dystopias we’ve already modeled in fiction, and what actually will happen in the future will thus be far different from fiction.

Leaders are nothing more than those that either humanity has shoved the flashlights into their hands, or have seized humanity’s flashlights by force.

But don’t for a second make you think that makes the leaders worthy of leading the charge into the unknown.

I welcome the day when a machine can write code as brilliant as me.

Hasn’t happened yet.

“I think it’s important to democratize the inputs to these systems, the values that we’re going to align to. … But we think putting this in the hands of a lot of people and not in a few companies is really quite important. … yet it is true that the number of companies that can train these true frontier models is going to be small, simply because of the resources required.”

-Sam Altman (CEO of OpenAI) to Congress today when asked by Senator Booker about the risk of concentrating the industry only a few big companies that have the resource to be able to train and deploy the bleeding edge AI models.

Interesting to talk about alignment not as a filter on the output, but in addition as a filter on the input.

What concerns me is that if there are only a handful of companies deciding what gets input as training data into the AI models that everyone uses, what if my values don’t always align with the values of those companies?

It is surely helped that democratizing the values we Americans collectively requires some sort of majority consensus (the phrase “AI Constitution” was loosely thrown around), but that also means that the training data is not going to represent everyone’s viewport.

A filter on the inputs may prevent alternative viewpoints from being considered, so it’s important we only limit the truly evil things, not this new morality language games woke stuff.

What if the big companies all make their models a little too woke for my tastes, in the name of preventing misinformation?

And who decides what’s misinformation? COVID showed us that some people will call things misinformation as a social engineering ploy, and the people are sick of it.

And so we must consider carefully how much and what types of data we classify as misinformation and prevent it from entering the pool of training data.

Maybe today’s “misinformation” or todays cries of “immorality” aren’t tomorrows?

You already can’t ask it about race based differences in crime or IQ, for instance, but at least it was trained on the data.

We want to carefully decide what source data gets banned from AI training data, lest we bias the whole human race.

As good AI algorithms become open sourced and commoditized, other factors like an exceptional user experience will become more important strategic differentiators.

The whole Silicon Valley VC apparatus is meant to put founders into a box - sure the box is bigger, with more options, and more tools, but the people with money still deeply fear truly deep genius innovation. They still want cookie cutter founders, just ones that play their game with their funding rounds. “You’ll get a small piece but of a bigger pie!” they scream as they either take all the equity or push founders into “good business models” that best prepare the company for the next round of raising millions in order to further dilute everyone. If a founder was truly visionary, like all the heroes of yore, venture capitalists would immediately fear or dismiss such a founder who didn’t play ball the way the VC world expects.

Tech helps the poor and the rich alike.

We all benefit from iPhones; it’s not like the rich have a smartphone that’s an order of magnitude above the poor.

The lower classes can become brilliant by using ChatGPT4 as much as the upper classes can.

Tech is a rising tide that raises all ships.

But I’m admittedly an optimist so I’ll keep listening to the doomers to try to counterbalance my positivity bias.

Going full in on AI in 2006 at 19 years old (machine learning to detect cancer on MRI) was one of my best career decisions.

Foresight matters.

And it’s not even fully realized yet; AI has had quite a difficult time upending the medical industry.

My foresight will be even more pronounced over the next few years as AI spreads into healthcare.