Many people in Big Tech appear to have lost their minds.

If you haven’t already heard, Jacob Coxon, an Anthropic engineer, publicly announced his resignation on social media, claiming generative AI companies are acting irresponsibly in their quest for superintelligence. On this point, I and many others, even those in the tech industry, agree. To understand the details of why recklessness on the part of AI companies is the core issue, not some mythical superintelligence, I highly recommend reading Cal Newport’s recent newsletter.

This kind of irresponsible behavior, in itself, shouldn’t too shocking. After all, Dario Amodei resigned from OpenAI himself in 2020 over similar concerns. He claimed he founded Anthropic, OpenAI’s top competitor in the U.S., in order to design AI systems with “safety and controllability in mind.”

Words that should haunt him. Because it turns out Anthropic’s own head of Alignment Science, Evan Hubinger, the very man responsible for ensuring models act ethically rather than just optimizing training objectives, added this to Jacob’s resignation thread

​Samuel Marks, a third engineer at Anthropic, felt the need to pile on, adding:

The idea that a large number of AI engineers and executives genuinely believe their product will destroy humanity within ten years has generated yet another round of media attention and hype (along with a lot of eye rolls from me). It’s not clear why, since as Melanie Mitchell, a professor at the Santa Fe Institute pointed out, these claims aren’t new either.

What is new is that several big names in the industry, including some AI founders like Gary Marcus, are calling for a boycott of generative AI, even as he makes clear the doesn’t buy into the hyperbole of human extinction. As he says,

I’m glad to see this happen. I don’t want to brag here, but I called for people to quit generative AI tools like ChatGPT and Claude more than a year ago.

As nice as it is to have some backup, those calling for a pause in research until the companies have a plan for better alignment are, in my opinion, missing the bigger picture. They are criticizing the approach these companies are taking rather than the technology itself. Ryan Calo, a law professor specializing in AI, theorizes why AI companies would sound the alarm on their own system in the first place (note that this is from 2023—ha!):

This sounds bad, right? Like we shouldn’t be using it, much less scaling it, regardless of the nebulous threat of human extinction.

It’s important to make the distinction here (as Cal Newport does) between large language models (LLM’s) and AI. Get rid of LLM’s and you still have a fully functional internet, movie recommendations on Netflix, and advances in predicting how proteins fold, among a host of other advances. I’m not suggesting we should stop all AI research/usage, just LLM’s and the awful data centers that power them.

You might argue that I’m throwing the baby out with the bathwater, that LLM’s are useful. But utility alone is not enough of a reason to keep using a tool. You can see the ridiculousness of this argument by substituting a different “tool”:

Still on the fence? Here’s a series of questions you might ask yourself:

• If you knew every query resulted in one person developing asthma, would you still use it?
• If you knew every query resulted in one person losing access to drinking water (or many people), would you still use it?
• If you knew that every query resulted in an activist getting arrested, would you still use it?
• If you knew that the data centers that LLM’s require were accelerating global warming, causing widespread disruptions to society and ecosystems, with the long term potential to render humans extinct, would you still use it?

It’s all well and good to talk about alignment, to complain that AI models don’t (and technically can’t) share human values and morals. But if we aren’t willing to live those values and morals ourselves, maybe the problem isn’t just the tech executives or the training approaches they decide to use. Maybe it’s us.

I urge you to embrace your power. Demand destruction—“a permanent or sustained decline in the demand for a certain good”—is a powerful tool that actually works. It sinks IPO’s and increases the chances of government regulation. It makes investment in data center buildouts less tenable. It puts pressure on business leaders to change policies and cancel contracts.

The steps to demand destruction are easy. Delete your accounts with LLM’s like ChatGPT, Claude, and others. Add “-AI” to your internet searches. Tell your friends and co-workers you don’t want to see AI content in your chats and emails. Make sure your boss knows that even AI insiders are calling for a moratorium on generative AI tools, making its continued use a real risk.

Remember… only YOU can prevent the harms of generative AI.

The tech execs are counting on you not realizing your own power.

​
Note: Before this breaking AI "news," I was working on a different newsletter, explaining why I think no one will be driving fossil fueled cars in 40 years. What I realized is that these kinds of newsletters take a tremendous amount of research to write. Since it’s just me, trying to share my thinking while still living my best life, I plan to switch to a monthly format. At least for now.