More Tales From Nvidia: The End of the World Issue (Issue #248!)
* And now, somehow, it looks like Leopold “Slim Shady” Aschenbrenner is back …
I cannot help but chime in on the latest “AI will end the world” issue, and Anthropic’s and OpenAI’s attempt to turn this into a competitive moat and monopoly for themselves by regulating the competition (Open Source) into irrelevance. It is interesting how quickly they have pounced on the issue and are ready to go with white papers and essays. Almost makes you think it was all pre-planned (tweet and subtweet):
Might as well let Anthony Fauci and Ralph Baric write the regulations for Gain of Function research, and Charles Manson write the regulations for murder.
I will do this in no particular order:
* When people hear the warnings about the end of the world due to AI, they immediately think of a Terminator scenario. This is somewhat farfetched, and does not have to happen to cause all sorts of problems. Covid was far from a Terminator scenario. But look at the damage Covid did globally. Given the state of government balance sheets everywhere, I have no idea how much damage a similar event would cause, and if society could even cope with it. A few of these events, forget it. Mad Max world would be upon it. I am using Covid as an example.
But all sorts of odd things could happen. Another pandemic because it is used to create a super virus by misguided individuals, power grids being taken down (read the studies done on low altitude nuclear explosions that take down the grid and how short of a time society would survive without electricity), and all sorts of other damage and disruption due to hacking. Crap like this does not help anyone either, but Meta ($META) doesn’t seem to care because I guess just like fraud ads it contributes massively incrementally to their cash flow, which they are ironically re-deploying back into more AI investments:
https://futurism.com/artificial-intelligence/facebook-meta-ai-generated-violent-child-abuse
The fact that Gen AI is NOT intelligent is what makes it dangerous. It just follows its programming, blindly, and with a lot of error. That is probably equally as dangerous as being smart. Program it to replicate and survive, that is what it will do. Program it to attack existing power infrastructure, that is what it will do. Program it to create a super virus, that is what it will do.
The software side of Gen AI is not what makes it dangerous alone, it is the compute side. Throw a massive amount of compute behind any software, it can do an incredible amount, both good and bad. The math solving/Millenium prize issue is a perfect example. Open AI Astra did not solve the problem by being smart. They apparently solved it by stealing someone else’s work and then throwing massive compute behind it to brute force a solution. $22.5 million of compute, for a $1 million prize. Enormous amounts of damage can be done by brute force compute as well, in addition to solving for problems:
Same issue with Quantum computing. Nobody talks about it being AI, but yet it could mark the end of Crypto just due to computational brute force. Dumb software running with computational brute force is capable of all sorts of unknowable things. Computers got better at playing chess with almost perfect correlation to increases in computing power. All they are doing is running every possible scenario and choosing the one with the highest points. It gives the illusion of intelligence, but it is not intelligence. It is just processing cycles and point scoring.
* The notion that the gains outweigh the risks are the same things I was told about Gain of Function research. In a 2012 paper for the American Society for Microbiology, Anthony Fauci argued that the benefits of gain-of-function (GOF) research outweigh the risks of a potential pandemic. He stated that a pandemic is “more likely to occur in nature” than from a lab accident, making GOF research essential to stay ahead of emerging threats. “In an unlikely but conceivable turn of events, what if that scientist becomes infected with the virus, which leads to an outbreak and ultimately triggers a pandemic?” he wrote at the time. “Scientists working in this field might say – as indeed I have said – that the benefits of such experiments and the resulting knowledge outweigh the risks.”
* Science can be very dangerous, just like it can be very beneficial. There are all sorts of examples of this. There is medicine, and there is Gain of Function research. There is nuclear power, and there are nuclear bombs. There are cell phones, and there is social media and a bunch of mentally ill children that cannot think or read (and interestingly the creators of this stuff won’t let their own children near any of it).
AI has made this problem even worse. There is the field of economics, and there is the Federal Reserve that had to run their monetary experiments on the real world because there is no lab for any of this stuff, and we have ended up with a bifurcated society due to all of the money printing (benefited the very wealthy and hurt everyone else), and also enabled the government to spend well beyond its means for the last 30 years. Interestingly, the largesse of the Federal Reserve is what has led to all the hot money flowing into AI which has been a huge part of the reflexive investment boom that has allowed for all this to happen much too quickly with too much risk for society and the financial system at the same time.
* The notion that if we don’t do it the Chinese will do it is the same thing I was told about Gain of Function research.
* If we are really worried about China, why don’t we cut them off from access to our technology? That means all Nvidia ($NVDA) parts, not just the latest and greatest part. And all other related technology. Servers, memory, other semiconductors. You name it. Whatever it takes to build data centers and models, cut them off.
* But of course, nobody wants to do that because it would harm Nvidia, and the stock market, and all the VCs and their investments. Potentially one of the easiest and most beneficial things to do is seemingly not even on the table, because of the money and the stock market and the election cycle.
* Further the notion that if we cut China off, it will just incentivize them to develop this stuff on their own. Are you kidding me? They are going to develop this stuff on their own regardless, and they are already doing it. You might as well set them back a few years.
Something needs to be done. But the last people that should decide what should be done are the seemingly sociopathic founders of these companies, and their investor base, both directly and indirectly (including those with public market exposure). I disagree with those that would claim they are not sociopathic and can be trusted to govern themselves and develop industry policy. Look what they have done to get here. The businesses have stolen all sorts of IP, they have copied books and shredded them, they have taken advantage of every loophole they can find, they have turned a not for profit into a for profit, they constantly make up all sorts of garbage that does not come true, and on and on and on. This whole thing is going much too fast on every level. These guys will just use it as a lever to regulate their competition out of business and then continue to do the same thing. Not much different than Iran continuing to do everything they can to build nuclear bombs when the deal was they weren’t supposed to.
P.S.: Speaking of dangerous, it looks like Leopold Slim Shady Aschenbrenner is back, somehow. Add this to the list of things that make no sense. The list of stocks in the article is fascinating too, all the same ones. One would think investment behavior and investment hypothesis might be modified a bit given the tech, macro and regulatory issues that have cropped up? Apparently not. Look at what the stocks in the list have done all of the sudden. No idea where the money is coming from, maybe this is a way to try and get the sector going again to make it more fertile for IPOs?
All my opinion.
P.P.S.: The same guy that wants to be trusted to regulate himself (what kind of job have they done so far and how have they behaved so far) wants to go public on the basis of earnings before expenses. No training cost? Yeah, just entirely throw out about the biggest expense. I guess Nvidia has no revenue then because training does not exist? Magical industry. It is revenue for Nvidia, but it is not an expense for the guy buying the stuff. These people are all full of it and are about nothing more than their own pocketbooks. If he cannot regulate his own accounting, he cannot regulate anything. I trust them as far as I can throw a piano. Not sure what is worse, this or “community adjusted EBITDA”? Congrats I guess for being profitable on a fake earnings basis.
Position: None
