---Disclaimer: I am not a financial advisor. I don't have any "Certificates". This blog post is offered for entertainment purposes.---
Background on AI's foundational technology (numerical methods)
In numerical methods, "integral based algorithms" are forgiving and robust while "differential methods" are rife with instabilities. Rates of change (X,t) tend to multiply measurement error. Acceleration of rate of change (X,,t) is even filthier. "Jerk", the rate of change of acceleration (X,,,t) is even filthier than acceleration.
Also from numerical methods, interpolation (estimating values that are bounded by measured data) is pretty safe while extrapolation (estimating values that are outside the cloud of measured data i.e., future predictions) get squirrelly very quickly. The farther into the future the prediction, the squirrellier the number.
This is important for two reasons. My AI expert informs me that LLM are basically "Auto-complete on steroids". They are guessing what the next word will be. At some point that runs out of gas. The other way it comes into play is that the astronomical valuations and ability to pull financing is based on speculation about how AI will fundamentally transform the economy like petroleum, semiconductors and the internet did.
Can any of the proponents of AI offer a credible guess as to when the venture will be profitable, covering both the costs of the sunk investment and the variable cost of the energy to run them?
Frankly, I think they are barking up the wrong tree. The AI that will be profitable will be tiny chips embedded in drones (unmanned, aerial vehicles) and will parse out potential targets and communicate with other drones in its cloud. Survival on the battle field will involve keeping your IFF helmet fully charged and the antenna undamaged and transmitting.
"But you HAVE to be investing in AI because that is where the stock-prices are exploding!!!"
I sort of am. 30% of my retirement fund (calm your beating heart...it isn't that much money) is invested in various equity index funds. Since NVIDIA, Oracle, MS and Alphabet are a substantial slice of the S&P 500, I am invested in them.
I am fine missing out on "beating the market". The exquisite agony of being "left behind" combines the two major forces in the market. It combines both Fear and Greed all in one package. It is fog-of-war and blindness-from-testosterone combined into one package and is virtually guaranteed to result in risky bets.
Errors AI seems to be prone to
"Nothing is better than God.Warm beer is better than nothing."
***apply transitive property***
"Warm beer is better than God."
Words can have very different meanings depending on context.
"You are a sight for sore eyes"
Oscar Wilde's original intention was "...a sight to cause sore eyes..."
"You look like the first breath of spring!"
The only survivable way to tell a woman that she looks like the end of a long, hard winter.
(From an AI generated Youtube video) "Alvin York charged the trench filled with 126 German soldiers armed only with his Springfield model 1903 and his 1911 Colt handgun"
While the "Standard" rifle for the U.S. Army was the Springfield model 1903, there were not enough in inventory to issue to the troops who were sent to Europe in 1917. Rather, they were issued the Enfield M-17 chambered in 30-06. In this case, "standard" and boots-on-ground reality were different.
Mountains of money are being sunk into "Data centers" and AI. I am humble enough to acknowledge that I may be very wrong. But I fear that the results will be more dystopian than empowering.
Energy build outs as well as removing all that heat from the easily damaged chips is kind of a limit.
ReplyDeleteBrownouts and blackouts in our grid have become a "normal" result of "extreme weather" (like snow storms).
Now we want Data Centers eat power like cities.
23.5 million gallons
The NSA's Utah Data Center has been reported to use over 23.5 million gallons of water in June 2022, which is significantly more than the amount of water used to fill the swimming pool at the Paris Olympics in just one month. This high water usage has raised concerns about the environmental impact of data centers in the region, particularly given the Great Salt Lake's declining water levels. Additionally, the facility is part of a larger trend of data centers in Utah consuming large amounts of water, with some facilities using up to 5 million gallons daily.
The Salt Lake Tribune
+1
Water issues in many states already, now let's use up MORE..
Gee Wally, what could go wrong...
Using up water? Where does the water go? is it polluted?
DeleteJust a little imagination could create some uses for pre-heated water, such as - warm it up a little more for steam to drive generators / air conditioners?
My ignorance of the subject is profound, but I do know that water is never "used up".
Stealing the auto-complete quote. I keep telling people that they are nothing more than 'smart' search engines with fancy I/O capability. They seek patterns based on frequency of proximity. From data that is at best average in quality - and that is actually a lofty goal given what they admit to using to 'train' their models. I'm too old to argue anymore - I just smile when they discover what I told them earlier.
ReplyDeleteYour numerical issues discussion brought back lessons learned painfully in math and engineering Numerical Methods courses. And later memories of investigating why model results were laughably wrong on inspection.
The real flyin the ointment is that the existing AI models “hallucinate”. In layman’s terms they make shit up…. When called out on this the models will apologies and then proceed to do it again! Apparently this trait is so deeply wired into the operating system that it really can’t be eliminated.
ReplyDeleteSo, we have an AI that we humans will have to fact check since creating an AI that is absolutely trustworthy in its accuracy is prohibitively expensive.
When you godown the AI rabbit hole it starts to feel like you’ve stumbled into a Monty Python production.
Some people claim that AI would make superior instructors for our children. Endless patience and no effects from insults from children. Information taught in person and recited or taught on a screen - no difference. Perhaps.
ReplyDeleteTeaching isn't just spewing information for others to learn. Teachers are part social worker (the good ones), who when confronted with a student who does not appear to be learning the material has to ascertain why that is the case. The home environment is noted - does the student submit complete homework assignments in timely fashion. Or are they always late and hurriedly written.
I don't see how AI can piece that together. And when that happens, that student will be left our. Not good.
ERJ, thanks for the early morning chuckle. First breath of Spring, indeed!
ReplyDeleteGIGO.
ReplyDeleteAlways thus.
AI gets its info from this internet thingy. Much of what's on this internet is garbage.
I trust AI not one tiny little bit.