The prevailing implementation of capitalism compels all companies to continue developing revenue streams to increase their overall “worth.”
Any company that has unique or rare data is compelled to do things with it. Those that don’t either can’t figure out how or explicitly reject the reward function of contemporary capitalism. We should really expect those deviations to be the exception.
When we let the market bubble-up protective conditions through buyer behavior, we advantage innovation at the cost of accepting more harms, because the market response is always reactive instead of proactive, and the reaction can sometimes take decades or more (like GHG emissions and global warming).
When we let structural regulations assert protective conditions on a market, we try to advantage proactive harm reduction at the cost of innovation, because artificial market limitations will be barriers to innovation and create secondary game conditions that advantage some players.
Which way we lean should depend on the type and severity of potential harms, especially with consideration of how permanent or non-reversible those harms are.
We also need to consider the confounding effect of corporate performance and recession expectations.
Cost centers in businesses are early canaries of expected pain, and a reduction in security roles may reflect belt-tightening irrespective of AI impact.
By avoiding fishing, you stop damaging many of the carbon sink systems in the ocean, and so as a second-order effect improve the sinks we used to have.
Bell also "just" provided telephone service, but had a nation-wide practical monopoly, and so was broken up into different regional operators.
The US has since gotten bad at dealing with companies like this. A company like Google or Amazon could/should be broken up into a few parts and would likely result in those parts together being worth more than when it was just the one company, and more competition in each industry.
Too early and the break-up can kill innovation. Too late and the company will have been a rent-seeking operation for so long that it chokes out dynamism.
It might be too early for OpenAI, but we shouldn't wait until they own all of the next Internet in the way Facebook, Google and Amazon ended up.
Yeah: it had a nationwide monopoly. Breaking up the Bells meant breaking that monopoly. That was the public policy goal of forcibly reorganizing the company.
How do you apply that to the OpenAI case? You have to draw the rest of the owl here.
No, I'm just not being clear enough. OpenAI doesn't have a monopoly. They don't even meaningfully have colluding BUs; you can't "break them up" (you'd basically just be killing it).
Killing or intentionally degrading a business can be legit public policy, but then just say that, don't pretend that the problem is that OpenAI is anticompetitive.
> It might be too early for OpenAI, but we shouldn't wait until they own all of the next Internet in the way Facebook, Google and Amazon ended up.
He wasn't proposing breaking it up as it is, which is what your comment assumes.
However, as I said in another comment, they do have other products (Sora and others in the pipeline that are meaningfully separate from their core product). I'd agree it's too early to break them up (at least by conventional anti-trust standards), and that if anything, they should be regulated on other dimensions.
If you claim, for example, that an input is not stored, but examples of internal steps of an inference run _is_ retained, then this paper may suggest a means for recovering the input prompt.
Any company that has unique or rare data is compelled to do things with it. Those that don’t either can’t figure out how or explicitly reject the reward function of contemporary capitalism. We should really expect those deviations to be the exception.
reply