I think the key question is: How can you be sure the supervisor/orchestrator agents are reliable? You are just pushing the complexity down into another layer.

You can't be sure but the point is you can be more sure, since agent 2 ("agent" which is really just a fancy way of saying some code that calls anthropics api) has only the context to look for a violation of a single rule.

Sounds like it was a prototype to validate an idea?

I think at validation stage technical details like that shouldn’t matter. All that matters is there market demand for this.

If yes, go and build it properly.

Sadly I don't think management would go and build it properly, this sort of thing happens frequently where the prototype is put directly into production because why waste time redoing something that already exists and works. Just got to clean it up a bit, round off some sharp corners, and put it into production post-haste.

This is true but not always.

Sometimes people would have enough time for a product tour and still skip it because no one wants to be forced to do anything.

While this might be true I’m worried about the hardware side of things.

What if you have a good enough model but the cloud model providers are better in procuring hardware for interference?

Local inference is definitely going to make more and more sense. Modern CPUs have all this amazing hardware well-optimized for inference purposes. I use a lot of web tools and see AI baked in and it feels weird. I want the smartness localized for speed and data security. I think and hope the industry points towards smart ai agents operating as locally as possible.

The cloud providers are probably better at procuring hardware for inference, but on prem users are better at repurposing hardware that they'd need anyway for their existing uses. In a world where AI compute is likely inherently scarce, it makes sense to rely on both.

I personally believe that eventually manufacturers will want to sell more of their hardware and look for ways to sell hardware to consumers. isnt that situation quite similar to the days of early computers? I am for sure biased in hoping that will be the case

Perhaps for some very specific capabilities such as TTS, translation, voice recognition and so on. But for general intelligence models, better hardware just directly allows better models and that doesn't seem to be changing any time soon.

I'm pretty sure that's not linear, so I personally expect the benefits of larger models to diminish. The question is at what point that's the case. I guess a lot of variables play into it, but it is possible that the benefits of running larger models will be too expensive for the little benefit they provide

You’ll be able to run the open models on any cloud at the cost of the hardware rental. While the closed models will try to mark up beyond the base cost.

If you’re interested how some of these things got build in New York in the past I recommend the books of Robert Caro about Robert Moses.

Building new massive infrastructure requires a level of ruthlessness that is not socially acceptable these days.

Op's example was underground. Moses built above ground, thereby requiring the ruthlessness. Not sure the same ruthlessness would be needed with tunnels.

According to Bloomberg[1] construction of the first phase of the second avenue subway cost about 2.5B USD per mile.

At that rate, even if you just look at extending the A/C/E from Jamaica to JFK, you're talking about 15B or so USD. And compared to today's [subway|LIRR] -> airtrain system, you probably only cut about 25% of the travel time (from 60 minutes down to 45 minutes)

Compare that to, for example, the Gateway Tunnel, estimated to cost about 16B USD and double the daily commuter capacity from NJ to NYC (including traffic to and from EWR!), and it's hard to justify new infrastructure to make it easier to get to the airport.

1. In NYC Subway, a Case Study in Runaway Transit Construction Costs - Bloomberg https://share.google/SPcN8iRDZG7lNiwt9

That’s something like $2k/resident to build, and then ongoing maintenance. Seems high for something that most will never use.

> Not sure the same ruthlessness would be needed with tunnels

Still requires lots of cut and cover due to buried power and water mains being poorly documented. And stations will require razing buildings, as well as gentrifying neighborhoods.

It’s not only about underground vs evicting people.

It’s also in large part about making sure that your project gets the required funding and other (social) projects don’t.

Further empowering and depending on either of those companies as a middleman in our lives should make us nauseous.

And this was probably also the prompt someone used to generate the article.

While I don’t want to downplay the toxicity of smartphones, couldn’t the same be said for books and newspapers?

When you are reading a book, you certainly need to use your attention. However, you stay in a given topic/world for a sustained amount of time. This feels very different and much less tiring than scrolling on your phone jumping from topic to topic. Especially social media feeds that have been optimized to keep using it as long as possible (dopamine hits and all).

Newspapers are probably an intermediate between those two, to various degrees depending on the specific newspaper (trash vs deeper analysis).

I think it can. In extreme cases (say grad school) I've had books and "book learnin'" suck the life out of my life.

"All I can do is read a book to stay awake And it rips my mind away but it's a great escape." - Blind Melon

I think reading is the difference. People didn’t whip out a newspaper when they had less than 30 seconds available. The smartphone has filled these gaps with an infinite amount of content.

Also, community. In a doctors office reading a paper - it is the same thing your neighbor is reading so you can talk about it. With smartphones, this is lost unless there is a pressing global event.

Yes, but you tend to carry around a smartphone all the time and the temptation to whip it out whenever there's more than a 5s window can be very strong.

A smartphone, at least with a connection to the internet, is always new. There is always something new to see and hear.

Arguably, same with books and - even moreso - newspapers. I vaguely remember doomsaying about people only scanning newspaper headlines.

But think about it, a good newspaper has a mix of news, background, entertainment, opinions, adverts, etc - not unlike browsing reddit or twitter, it's a barrage of emotional ups and downs and items asking for your attention in different ways.

With that in mind I don't think the concept of distraction is new.

Once you read thru the newspaper you are done. Nothing new left. Not so with the smartphone. Always new.

The both have an end and limited novelty.

Give it hard enough problems?

I think, Spotify is being optimized for attention KPIs like time spent in app. If you’re listening to music, you don’t necessarily need to look at the app.

But with videos…

If you’re in the Apple ecosystem, Apple Music is the obvious alternative which gives you the plain experience. If not, maybe Amazon Music?