221–230 of 363 posts
Predictive maintenance != “automating a CEO”
CEOs aren't paid for the day-to-day value they provide to the company. They're paid for the long-term, strategic value the owners think they might provide the company in comparison to other candidates. So if you think CEO Candidate X will make you a 3x return on your investment, but CEO Candidate Y will make you a 5x return, it's almost certainly worth it to pay him/her the extra $200K to get you there, if that's the…
It is naive to think that every other position within a company could be eliminated except the CEO. Of course it could be, and eventually you can expect that it will be. It's when humans are no longer the shareholders that we all need to be concerned. If we're still alive.
Earlier quoted context omitted.
No he isn't.
I think you might be misunderstanding what "All numbers in thousands" means on that table she linked. EDIT: To be clear, Walmart's revenue in 2020 was 559 Billion with a B[0]. We can see on that table[1] that if we take the values listed in the "Total Revenue" row (559 Million with an M, 523 Million with an M, etc) and multiply them by 1,000 according to the directions ("All numbers in thousands") on the chart that w…
It's casually flippant to say that all CEOs must be automated out of their jobs without looking at the details and nuance of what the data bear.
So much "interesting stuff" both in the article and in the comments (my favorite is the guy claiming he knows of multiple startups where the startup CEOs are making 7 figures). I was initially perplexed when I saw this article doing well on the HN homepage, but I assumed that it's just some kind of an engagement metric that the HN algorithm is picking up on. Really disheartened to see that the article itself and so m…
Maybe that was the case several years ago, but I believe most people on HN are regular salaried employees without any particular startup ambitions.
Earlier quoted context omitted.
How are resources allocated in this utopia?
My personal guess is - assuming that democracy is in tact, redistribution will be voted in with landslide support. In a hypothetical future where almost nobody can compete with an AGI and 90% of voters find themselves completely useless in the economy, the proportion of voters who are against redistribution will drop precipitously.
Everyone is allocated the same by the state? A society where no-one has any function to society and where the state allocates resources to individuals sounds more dystopian than utopian to me...
Earlier quoted context omitted.
There was a Danish study that showed that a death in the family of a Danish CEO led, on average, to a 9 percent decline in the profitability of the corporation. If it was the death of a spouse, the decline was 15 percent and, if it was a child who died, 21 percent. From the study: “Interestingly, similar deaths experienced by individual members of the board of directors do not significantly affect firms’ outcomes. Ou…
To be honest, this shows that a dysfunctional CEO (for completely understandable reasons) leads to a nonprofitable firm, not that value is added. You could say that people with no or defective keyboard have a hard time programming, but the best keyboard does not add significant value to a developer.
Earlier quoted context omitted.
>You need profits for stock buybacks. You merely need capital. Lots of buybacks are funded by debt
This is pretty insane actually. Is there a common metric that captures which companies are funding buybacks with debt? or is it easy to figure out? I am not a finance person.
1. Specifying the problem to be solved can be very subtle. The article throws in an example of scheduling maintenance in a transit system. That's a case where someone can specify some utility function around service disruptions, cost, etc. Companies are complex.
2. The space of actions available to a company at any point is extremely broad, and also hard to specify.
3a. An agent needs to learn from something. This can take a couple directions. Roughly, "off-policy learning" learns from the actions and outcomes of _another_ agent with different internal rules etc. Perhaps we have incomplete data of this sort (business school case studies etc leave much information out). But you need a _lot_ of data, and it needs to cover the same sorts of situations that your agent would get around to. And some actions probably the software agent _cannot_ copy (e.g. doing an interview on Sunday news program).
3b. "On-policy" learning is improving your agent from the past actions/outcomes of its own decision-making. Normally this means we need some simulation environment in which to let the agent learn. How do you simulate the whole economy, and market forces with which a company interacts? After learning in a simulated environment, you likely need to further improve (and test) in the real world. In this domain, I think that means you have to be prepared to let a computer program run many companies into the ground.
4. Without a decent model of the world around them, RL agents have a harder time attributing outcomes to specific past actions. You beat your earnings estimate, but which of the bajillions of decisions made over the past several quarters is most responsible for that? I think human CEOs actually do a questionable job of this, but an automated one would be worse.
Earlier quoted context omitted.
>You need profits for stock buybacks. You merely need capital. Lots of buybacks are funded by debt
This is pretty insane actually. Is there a common metric that captures which companies are funding buybacks with debt? or is it easy to figure out? I am not a finance person.
https://fortune.com/2019/08/20/stock-buybacks-debt-financed/
https://www.cnbc.com/2019/07/29/buybacks-companies-increasin...
https://crsreports.congress.gov/product/pdf/IF/IF11393
From the last one: "Data on the percentage of buybacks that are leveraged are inconsistent. For example, the bank J.P. Morgan Chase reported that leveraged buybacks accounted for 14% of overall buybacks during 2018, which it said was the lowest percentage since 2009. But Yardeni Research, a respected securities market research firm, reported that they were 56% of the total."