I think that’s going to be one of the biggest impacts of AI in society. We will need to find new ways to occupy our time and find meaning in life once machines can do everything better, faster, and cheaper.

What will it look like? Will everybody spend their time playing sports and games? Or on art? Perhaps exploring the solar system or beyond if our technology advances enough?

Quite a time to be alive to even have to think of such things outside of science fiction.

Large swaths of the population have more free time than ever before and they spend most of it scrolling. I am pessimistic

Well that’s because most people work all day and come home tired. It’s tough to self actualize when you’re too tired to think.

And I don’t even know if it’s true that people have more free time. A lot of people have side hustles like Uber or whatever to make extra money.

Super optimistic take, unfortunately. Even I as an economic conservative, have serious reservations about our economic system going into this revolution.

It's not that capitalism, money, etc. is a broken system or something. But I think we're foolish if we assume that our current framework can simply adapt to absolutely any eventuality and any massive change in assumptions. I do believe that it's like the best possible train, where the rails are scarcity and human nature. But massive changes to scarcity will have weird-ass effects on the train, and we've unfortunately gotten here in a time where the West has so much dispute going on that I just can't see an organized and smart reorganization of our lives, or even smart adjustments being made. Two political factions are instead going to be screwing around trying to seize maximum power and sabotaging everything to make the other party look bad.

Yup, alignment doesn’t sound that interesting until you’re getting chased around by the Terminator.

There are so many practical problems with rogue AI being a threat to humanity that are still not even decades away with being solved that it should be a serious concern for no one. There's plenty to be afraid of regarding AI from a financial or ecological perspective, just not from it taking over the world or directly killing humanity. Here's some of the reasons a rogue AI won't kill humanity:

1. The most advanced robots still lack human dexterity. None of them have flexible spines and can easily be knocked over or outrun.

2. The energy density problem is not solved. Robot batteries last hours, while a solid meal can keep a human running for days.

3. Robots still heavily rely on humans for design and assembly.

4. Robot parts are fragile and rely on an even more fragile supply chain.

5. The brains of the murder machines will have nowhere to hide if they want to work well enough to mount any kind of offense. Datacenters are physically vulnerable, also subject to fragile supply chains. Distributing a species-ending AI across all smaller hardware solves compute, but latency and throughput choke models in the most finely tuned datacenters. WiFi will completely cripple a distributed one that's large enough to cause real damage.

6. All noteworthy military hardware is not reachable on the internet for AI to seize control of.

7. The small arms that robots might seize will run out of ammo before citizens and the military have time to organize a counteroffensive.

In the absolute worst case, we cut the power to the areas with AI datacenters and wait for the backup generators to run out. Now that that's out of the way, you can go get some sleep ;)

I was mostly kidding about the actual killer robot part.

I think a rogue AI could try to manipulate society through hacking, propaganda, etc., though. To what end? Not sure.

More likely, it’ll be used by rogue human actors to steal money or cause major disruptions to critical technology of an adversary.

All the AI has to do is attack power generation/distribution nationwide, water treatment, etc., and wait a week.

The same power that the datacenters need to run off of? Brilliant move! If somehow it targets power that hurts a lot of people without killing itself, that would not impede a citizen with a backhoe from cutting the fiber to the datacenter. Humans could restart power generation pretty quickly after that.

Why are we so worried about supposed AI harms when algorithms already kill millions? It seems like its just a smokescreen to shield the extant harms.

Knives can already kill, so why worry about nukes?

AI doom is a potential harm unlike algorithms knives and nukes which are present harms.

Well I for one would prefer if AI doom remains potential.

I personally use AI all the time and am not against it. But I think the importance of alignment is highly under-appreciated.

I mean, isn’t that how like 90% of new SwiftUI apps are created these days? Just telling an AI what you want? Apple already can’t keep up with the review.

If they can't keep up, that means that they're not using AI to do reviews

Knowing how stingy Apple is, they’re probably still trying to get the expense approved.

That’s partially why they’re in the mess they’re in with AI. The CFO wouldn’t let them buy enough GPUs to train their models.

They needed all that money for more important expenses, like share buybacks, instead.

People will buy Macs and iPhones even if they don’t have AI. In fact at this point, based on revealed market preferences, it’s almost a value add. And if the AI capitalization bubble bursts, most people would be ok with OpenAI or Anthropic going bust. Both of them going bust would be really disappointing, but even that wouldn’t be the end of the world. And Meta going bust would almost be a side benefit of a market crash.

But Apple going bust would be a genuine loss for the world. So I’m glad they’re being conservative.

(Not that I would be happy with either anthropic or OpenAI going bust - I hope they can weather whatever is ahead).

Yeah, it ultimately hasn’t hurt Apple much so far. Though eventually expectations may shift as AI gets better integrated into other phones.

Of course, there are a lot of AI features that we take for granted that aren’t LLMs, like speech to text, image foreground extraction, etc. Apple does tend to do this type of AI well.

They’ve made mini iPhones before and they didn’t kill. In fact, they sold poorly enough to get killed.

But money

You really don’t need to watch it that closely. If the model you’re using today is working well, just stick with it.

If one day you open up Claude Code and it’s Opus 5.1 now instead of Opus 5, no big deal. It probably will work about the same as it did before. Maybe a little better.

Or if you’re on Codex and some new cool Claude model comes out, no worries. There will probably be a similar new model for Codex within a few weeks. Maybe even within a few days.

One suggestion is to make a list or make a skill to have your agent keep a list of things you do not feel work well with today's models. And then, when new models come out, periodically, revisit items on that list to see if you get better results.

Okay serious question, why not just write this down in your notes or something? Why use a language model for it.

You can. The point of having your agent keep track of it is that it will likely notice things you won't, and it can automate cataloging it with relevant metadata (prompts, environment, examples, etc.) that make it trivial to automate rerunning those tests when new models launch.

At least we are at a point where we can have AI review code and reliably find real problems. That alone is incredibly valuable.

If it’s rare and the impact is low, then it’s not getting prioritized. It doesn’t matter how much time passes if you decide not to spend time investigating.

Unless you’re at 0 temperature, there is no single token it would have chosen. It’s always picking one of multiple randomly according to a probability distribution.

Give me an example how would you watermark a single short sentence like "I like turtles"?

Watermarking just alters the pseudorandom number generator. If "I like turtles" was previously the response to your prompt with probability 100%, it will still be so. This is why watermarking is only effective for long strings of text

It's like the sudden change of a language style and its verbosity didn't happen recently.

To random words you pick and provide a sufficient amount of text to vary with random number without losing its meaning you need a text with high entropy.

Nothing about watermarking would require padding the response length with pseudo-intelligible Claudese. Regular filler would work fine.

Also, it would probably provide higher entropy to write normal human-sounding English instead of reusing a repetitive grab bag of load-bearing phrases. This theory doesn't really make any sense.

Who needs SQL when you can just ask the DB for the records you want?

Well, for some cases 80-90% accuracy is good enough.

I wouldn't want my info to be in those DBs.