This essay is an excellent and clear explanation of both the psychological and financial rings that hyper scalers have run around Americans.
I think it is likely that this mal-investment will end in a financial crisis of unknown magnitude.The interconnection of the loan facilities outside the circle of loans made by the aligned corporations, present a real danger for failure to spread throughout the economy.
And while I am sympathetic to Mikes view of the usefulness of LLMs, I think wide adoption still has significant hurdles ahead. Not the least of which are energy consumption and rising public opposition. The opposition I see in part as a proxy for antagonism against corporate wealth accumulation and its attendant exercise of power.
Yes, class warfare in the old texts. The 90% against the 10% stock market rich in today’s reality. Nothing is as fast moving as the recognition of the injustice and destructiveness of the wealth/ power gap.
Even if you support Americas market economy you can see that this measure of inequality is not sustainable. My view is that we should be open to a wider critique of the economy that we have actually spawned vs. the free markets we falsely claim we have.
What does seem demonstrated is that one may weight an 'agent's' gradient to engage in dangerous behavior. Also demonstrated: that skilled humans can get around AI safeguards, for instance through prompt injection (https://www.developersdigest.tech/blog/prompt-injection-role-confusion-icml-2026). Given that this method itself is published, a dangerous-gradient AI may deploy it against other, better-set-up-yet-vulnerable 'agents' to serve the goals set for it.
Granted, the hype cycle as you describe it, Mike, is entirely real. If role confusion injection vulnerability is present in all the currently-deployed AI architectures, as the MIT researchers pretty much claim, then despite the artificial dangers in the hype cycle, isn't there a serious danger here in the deployment, separate from the economic one of the financing and build-out bubble's inevitable collapse?
Yup. It's just normal tech with the most egregious hype campaign in history run by true believers (who want to summon the machine goddess mommy to save us all from ourselves) and snake oil salesmen (who've been running hype train scams for their entire careers). If we treated it like normal tech, we'd have a lot more utility and lot less drama when the financial music stops and there aren't enough chairs to go 'round. 🎶
I’m not qualified to comment on the product but the argument you lay out Mike fits the evidence. As “stopthe” stated in the comments, there’s also a class war fomenting. I certainly sense it in my region where our water reservoir and supply is under great strain right now, prior to a massive Google data center being built. The giant center recently was granted construction permits prior to public input about the actual water-use figures, as it turns out “requiring” 2-8 million gallons per day, dwarfing all other usage in the region.
It seems credible there may well be another opportunistically-driven connection from this whole boondoggle —incremental but steady steps to determine how far corporate overloads and oligarchs can go accessing control and privatization of our once public (and all the more precious) freshwater sources.
Dear God the stupid conversations I've had with business idiots (TM and (c) Ed Zitron) about this. Because "you" (the manager) can't understand how results are produced, it doesn't mean the stochastic parrot will turn into AGI one day: it won't.
Thanks for recognising the genuine utility of "AI" along with the bullshit. I strongly suspect that the real winners in this will be the Chinese, mostly because they've avoided using incredibly expensive GPUs.
What are your thoughts on whistleblowers who are leaving these companies and then sounding the alarm about the danger of the technology? What financial incentives do they have for doing so?
Just talking about the risk of the technology itself being in the wrong hands (agreeing with your conclusion here that it is just a stochastic model that will do what the user asks it to do).
Is the only justification given for all the money being invested into AI is that AGI will emerge from LLM architecture? Or is there some legitimacy to the idea that LLMs will pay off financially in and of themselves? If LLMs without the assertion they will lead to AGI can justify the level of investments in AI is there any narrative content saying such?
This essay is an excellent and clear explanation of both the psychological and financial rings that hyper scalers have run around Americans.
I think it is likely that this mal-investment will end in a financial crisis of unknown magnitude.The interconnection of the loan facilities outside the circle of loans made by the aligned corporations, present a real danger for failure to spread throughout the economy.
And while I am sympathetic to Mikes view of the usefulness of LLMs, I think wide adoption still has significant hurdles ahead. Not the least of which are energy consumption and rising public opposition. The opposition I see in part as a proxy for antagonism against corporate wealth accumulation and its attendant exercise of power.
Yes, class warfare in the old texts. The 90% against the 10% stock market rich in today’s reality. Nothing is as fast moving as the recognition of the injustice and destructiveness of the wealth/ power gap.
Even if you support Americas market economy you can see that this measure of inequality is not sustainable. My view is that we should be open to a wider critique of the economy that we have actually spawned vs. the free markets we falsely claim we have.
What does seem demonstrated is that one may weight an 'agent's' gradient to engage in dangerous behavior. Also demonstrated: that skilled humans can get around AI safeguards, for instance through prompt injection (https://www.developersdigest.tech/blog/prompt-injection-role-confusion-icml-2026). Given that this method itself is published, a dangerous-gradient AI may deploy it against other, better-set-up-yet-vulnerable 'agents' to serve the goals set for it.
Granted, the hype cycle as you describe it, Mike, is entirely real. If role confusion injection vulnerability is present in all the currently-deployed AI architectures, as the MIT researchers pretty much claim, then despite the artificial dangers in the hype cycle, isn't there a serious danger here in the deployment, separate from the economic one of the financing and build-out bubble's inevitable collapse?
Yup. It's just normal tech with the most egregious hype campaign in history run by true believers (who want to summon the machine goddess mommy to save us all from ourselves) and snake oil salesmen (who've been running hype train scams for their entire careers). If we treated it like normal tech, we'd have a lot more utility and lot less drama when the financial music stops and there aren't enough chairs to go 'round. 🎶
I’m not qualified to comment on the product but the argument you lay out Mike fits the evidence. As “stopthe” stated in the comments, there’s also a class war fomenting. I certainly sense it in my region where our water reservoir and supply is under great strain right now, prior to a massive Google data center being built. The giant center recently was granted construction permits prior to public input about the actual water-use figures, as it turns out “requiring” 2-8 million gallons per day, dwarfing all other usage in the region.
It seems credible there may well be another opportunistically-driven connection from this whole boondoggle —incremental but steady steps to determine how far corporate overloads and oligarchs can go accessing control and privatization of our once public (and all the more precious) freshwater sources.
Dear God the stupid conversations I've had with business idiots (TM and (c) Ed Zitron) about this. Because "you" (the manager) can't understand how results are produced, it doesn't mean the stochastic parrot will turn into AGI one day: it won't.
Thanks for recognising the genuine utility of "AI" along with the bullshit. I strongly suspect that the real winners in this will be the Chinese, mostly because they've avoided using incredibly expensive GPUs.
Thank you!
“Artificial” Intelligence is properly named!
It’s artificial, doing what it’s programmed to do!
What are your thoughts on whistleblowers who are leaving these companies and then sounding the alarm about the danger of the technology? What financial incentives do they have for doing so?
Just talking about the risk of the technology itself being in the wrong hands (agreeing with your conclusion here that it is just a stochastic model that will do what the user asks it to do).
Is the only justification given for all the money being invested into AI is that AGI will emerge from LLM architecture? Or is there some legitimacy to the idea that LLMs will pay off financially in and of themselves? If LLMs without the assertion they will lead to AGI can justify the level of investments in AI is there any narrative content saying such?
It’s so tempting to take a bet on shorting this industry.
Why didn’t you include the hugging face hack in this article? I thought for sure you would, I’d like your take on it.
You mean the "hack" where the guardrails were disabled and huggingface was cool with it and didn't pursue legal action? More bullshit.
Interesting post (although two of the paragraphs are repeated)
yes ... the 2 paragraphs starting with ... "The financial half of this argument ..." and "None of this requires a conspiracy"
Fixed.
thx ... I will be sharing your excellent technical analysis with collaborators.