Silicon Valley’s elites can’t be trusted with the future of AI

You're touching on the separation between the people who can and the people who cannot but wish to control. You get a lot of people bleating about AI, unabashedly with their own ideology behind it, but they tend not to be in the crew who can make whatever we are calling AI these days.

Largely agree, but it's not clear at this point who will control the AI future. SV elites today are like NYC/NJ/Philly elites of the 1960s. They have the momentum and the capital but the cost of doing business, red tape, and wages are too high. It's not clear they will control the next tech revolution.

It was inevitable that it would fall into the hands of consumers (leaked or released), so it's better that we deal with the consequences rather than putting it off.

The article seems like a vehicle for the author’s ethno-narcissism.

I saw that interview, Altman said (paraphrasing) "you won't be able to compete, there's no point trying". I didn't interpret it as directed at the example the questioner gave (a few engineers with a few million dollars), I took it to mean "we're so far ahead of you, you'll never be as good).

The arrogance was breathtaking, particularly when it's inevitable that OpenAI will be knocked down a few pegs (arguably, it's already happening).

Very true. But--Economic scales and incentives are artificial and must be tuned. For example, the USA has a badly regulated medical system that refuses to train doctors efficiently, so medical care is overpriced. 18 year-olds who want to become doctors go to college for like 36 weeks a year with maybe four or five hours of instruction a day. Massive waste of human potential. The legal system is very bad as well. Two graduates from different schools with the same score on the bar receive compensation varying by a factor of 3x dependent on where their degree was obtained. Going to Harvard law is indicative of having connections and the cognitive potential to engage in lucrative schemes well into the future, not actual skill. This does not happen in a country with an efficient legal system, like Germany.

I can tell you my personal reasons for thinking the concern of misinformation is valid: LLMs offer a tradeoff between time to generate content and credibility. You can generate ten thousand websites worth of misinformation in a day, at the expense of having it sound rather clunky, machine-ey. Sophisticated minds won't fall for it the way they might fall for savvy, hand-crafted propaganda articles. But recent experience seems to have shown that clunky webpages with clunky content is enough to influence a critical mass of low-sophistication citizens, polarize their views, make them very angry, shut down their ability for dialog, etc. That's concern #1.

Factoring in the enhancements to LLM output quality (which translates to credibility) we're likely to see in the very near future is concern #2: misinformation, mass propaganda, targeted propaganda, bogus news articles on fake websites, perhaps even crafted opinion pieces that can get through the gates and published onto legitimate news and commentary platforms, or full-blown legitimate-looking bogus news websites, all that kind of stuff will be available by the terabytes at the push of a button. No need for a bright, educated mind to sit at the typewriter, no need for an operator to even have a basic command of the target language.

All of this falls in the class of willful misinformation/disinformation/propaganda; reading your comment once more makes me think you might be wondering more about accidental misinformation, is this the case? I can already put forward this much: an immature or insufficiently educated person is unable to cause kind of harm to themselves or their surroundings with a typewriter compared to what can happen when you put an LLM-powered chatbot in their hands. I've heard of, or witnessed a significant number of people in their twenties, be them students or curious-minded guys, who would use ChatGPT as an authoritative resource upon which to base their work. It blew my mind to see this whole new way for not-unintelligent people to act willfully naive, whereas the appearance of the printing press and typewriters was accompanied (mostly) by an ethos of critical thinking, an ability for autonomous questioning and doubt. And if this wasn't the case in the early stages (you might even argue we're somewhat still in this early stage given the relationship of millions of Westerners to the work of toxic fiction known as the Old and New Testaments aka the Bible), then the case for being wary of what might come out of a printing press/typewriter and being concerned about its effect on society then was a valid, warranted one. What you funnily but derisively describe as to "regulate this machine and force its manufacturers to add features to the mechanism so that it cannot make hateful or erroneous sentences" has more realistically been conceived along the lines of: "we need to establish a basic hierarchy of knowledge, as well as a solid epistemological foundation accessible to people at large, a caste of authority figures / educators (clergy, professors) to guide young minds into this new realm of textual abundance, cultural guardrails against the really toxic stuff that might come about, etc."

The actual constraint is that AI has a fundamentally parasitic reproductive strategy, it cannot reproduce on its own and it's reproduction is likely to be detrimental to the host.

A sufficiently powerful AI will overshoot and kill it's host so to speak, in any number of conceivable scenarios that don't necessarily imply also killing all humans, just digital infrastructure.

This would have serious consequences, but those consequences probably would not include complete extinction for us, and the unsustainable nature of digital infrastructure was obvious way before AI advancements became a serious mainstream issue over the last several years. It's possible AI does not even represent a significant acceleration towards that end that was always going to happen.

So I think we are safe at least in that very limited sense.

I think that’s the definition of ad hominem, doesn’t address the points made in the article. As someone without strong opinions on this topic I appreciate hearing the various sides.