That said, AI is spiky, and I fully expect there to be a class of problems that AI is really good at solving and a (maybe larger) class where AI is worse than human mathematicians.

And even Trump's corruption is not the problem. If that were his only sin the US would be fine.

Our political disfunction is caused not by personal failings (although those abound) but by genuine disagreement on which direction the US should take.

But, IMHO, the difference between writing programs for fun and writing programs that other people find valuable is like the difference between masturbating and making babies. The former is fun, the latter is fulfilling.

There is nothing wrong with having fun. But if you ever find yourself unsatisfied with all your half-built projects, maybe consider writing software that other people value or even love.

Yes, it can be frustrating to try (and often fail) to satisfy others. But seeing someone enjoying your software is an amazing feeling. It makes you feel proud, useful, and talented all at once.

For Google I meant L8, which I guess is Principal Engineer now?

Another thing: People at FAANGs usually ride the stock up and get way more than 7% per year. I'm embarrassed to say that I sold a lot of MSFT stock at $30 in 2011. Today (~15 years later) it is trading at $530.

Still, if you reach staff-level at 40 years old, you could have $25M at 55. That's pretty amazing. But that requires lots of luck. Most people never make it to staff-level. I never did.

   A. Guaranteed $1,000 each month
   B. Every month, flip a coin. Heads gains $5,000, tails you lose $1,000.

To me, that's SPY vs. QQQ. Tech in general has higher expected return, but more volatility. If you can handle the volatility, then QQQ is better. If not, then not.

Ultimately, I'm taking a bet on tech because I believe tech has the greatest chance of improving the economy/world.

[I'm also not qualified, so do your own research]

If you cannot have a happy life on that, well...

But I get it--if your goal is to accumulate more wealth (no judgement--that's a legit goal) then it gets harder. When I was at Microsoft, partner-level engineers were making $1 million per year; now it's probably $2 million. Staff-engineer at FAANG is probably similar, I bet, to say nothing of the AI labs. I bet you could reach $25 million after 10 years with the right investment strategy. And $25 million throws off at least $1.25 million per year if you invest right.

Beyond that, you probably do need a startup or some other significant ownership stake.

The tricky part is that a lot of this advice is situation-dependent. Should you switch to a riskier company that offers a higher compensation? Sometimes yes, sometimes no. Should you work late to accomplish goals hoping to get a promotion? It depends.

So how are we supposed to know what to do?

The answer, in my opinion, is people. You have to trust people who are trustworthy and avoid people who are not. As an employee, your success depends on the people who hired you--they are the ones who can promote you or cheat you. The best leaders are the ones who realize that helping you advance is going to help them advance. Help your boss accomplish their corporate goals and they will promote you--not as a reward, but as a bet that you will help them on the next goal.

Bad leaders are either too selfish (fail to promote you) or too incompetent (unable to rise, even with your help). You need to learn to avoid them.

Therefore, my advice is this: Look at the leaders in your company--people who could plausibly hire you. Which ones are the best bets? Which ones have both the talent and the integrity to succeed? If you're lucky there's more than one and you can work to get on their team. If you're unlucky, there are none and you need to find a different company.

Yes, exactly. The issue isn't, "we don't know how to sandbox an agent", the issue is that they need to test the agent with full internet access because that's the product they are selling. They are selling a product that does things for you on the web. You have to be able to test it with a real web connection.

Maybe the answer is to simulate the web. What if you ask an agent to reproduce the key parts of the web (travel sites, government sites, etc.); effectively, clone a real, working version of the web, but inside a simulator. Now you could test the agents inside the simulator. Since "coding is a solved problem" maybe that wouldn't be as hard as it sounds. Right?

The argument from the frontier labs isn't, "We're worried that AI is going out of control, please stop us" it is, "We're taking a lot of risks to compete, so please limit competition."