ithkuil

Is there evidence that frontier models at anthropic, openai or google or whatnot are not using comparable optimizations to draw down their coats and that their markup is just higher because they can?

That entirely depends on the location of the proxy and the extra conditions you can express. E.g. you could bind it to a source IP and have the proxy check that, or use some overlay network (like tailscale does)

My point was that you don't literally have to run the proxy on localhost in order to scope the request.

The auth between your app and the proxy can be scoped more easily.

For example if the proxy runs in localhost you can trust the localhost workload.

Or you can use some other kind of workload identity proof (like cloud based metadata servers). If you leak such a key no other VM can use it, because it's scoped to your VM

in exchange for service that presumably a) costs something to amazon to operate (so not pure 100B profit) and b) anthropic would have to spend anyway to operate their business.

so basically ...

you could view this as a kind of discount, but instead of paying less later, you get some cash now and then pay full later.

Name checks out

If such tax would introduce an asymmetry that will favour human employment, then there should be enough buying power to create some demand

I wonder what percentage of the job space truly depends on the current edge we have over machines.

I think it's reasonable to worry that way before machines are more reliable than the average human (let alone more reliable than a highly trained human) they can pose a significant disruption to the job market which will send shockwaves throughout society

Consequential enough that it's reasonable to plan for it.

Just train machines on the huge corpus of human needs so they can need things like no human has needed things before.

What can possibly go wrong

Perfect place for the fortress of solitude!

Yeah. Basically conquest is possible when the victim is weakened. There are many ways to become weakened. Infighting and disease are common causes of weakening.

And indeed the Nahua people had writing.

It's not about where you come from. It's about whether writing is useful or even required for some aspects of civilization to develop.

You make it sound that there are only two sides in this story.

Spain, Argentina, Kenya, Indonesia, Kuwait and countless other countries haven't bombed any civilian infrastructure either and yet they will be affected by the aggressive posture around international maritime traffic.

Are you expecting that Iran will not apply the fee to ships that sell oil Malesia or South Africa?

The spec can be wrong for many reasons:

1. You can write a spec that builds something that is not what you actually wanted

2. You can write spec that is incoherent with itself or with the external world

3. You can write a spec that doesn't have sufficient mechanical sympathy with the tooling you have and so it requires you to all spec out more and more of the surrounding tech than you practically can.

All of those issues can be addressed by iterating on the spec with the help of agents. It's just an engineering practice, one that we have to become better at understanding

Yes, unless people had some incentives to show an increase in the trucking capacity in order to meet some metrics and get more funding etc. (not saying that's what's happening, but just as a counterpoint to your logic)

I wouldn't mind if they were actually competent in what they do.

yeah it actually works to use claude to reverse engineer itself; I've used that to workaround some problems. E.g. that's how I discovered that I had to put two slashes for absolute paths in sandbox config. The thing is, the claude team is so quick that soon enough they add more and more features and fix more and more bugs that your workarounds become obsolete

1. Yes this configuration applies to the sandbox where the commands executed by Claude are run and as such it applies to anything these commands do, including child processes etc

2. The sandbox rules also apply to the program written by the agent IF you ask Claude to run that program. If you run it manually from another she'll or via the "!" directive from within Claude, the sandbox won't be used

The sandbox only limits what processes spawned by Claude can do. Claude itself can read from any directory you tell it to read from (i.e. that's a different permission mechanism)

Not if the sandbox rule forbids reading the private key and the ssh agent socket (as the shown example does)

"All sensors are temperature sensors, some measure other things as well"

Well, that's why it's a good idea to have a place to try things out, so we can learn.

But of course, doing that is highly impractical for many other reasons.

The only advantage of terraforming Mars is that if you do it wrong you're not making it worse for anybody that lives there. It could be a good test bench if it wasn't for the elephant in the room: it takes a very long time to terraform a planet

That's unsurprising given that a lot of our own abilities as humans come from having painstakingly acquired practices and methodologies and tools (like pencil and paper, note taking, let alone algebra, formal methods and electromechanical aids). We call this "education" but it works in a way that is more similar to agentic harnesses than to pretraining or fine-tuning. This is reflected in the fundamental different way in which children and adults learn new skills

The harness would be fine if the agent coded its own harness in a controlled environment while observing the game.

Not sure if the specific rules of this prize allow that, but I would accept that

Also used in Italian and presumably in many other languages.

Like with any word, it's use in colloquial form may vary from generation to generation, from subculture to subculture etc

We burn the spice

But that handicap is due to domestic politics which is a real world force that shapes the american war machine ability more than anything

Even a short surprise can be crucial in an ancient battle, where breaking formation can be fatal

Perhaps it's a timing issue? Perhaps he would have time for more work but the calls cluster when he's busy?