I assume one can't benchmaxx multi-year long efforts, clean architecture, taste etc as easily as these "make tests pass" tasks

As prediction markets already show, forecasts can influence the outcomes they are trying to predict.

What happens when these models become extremely accurate and widely trusted? A forecast like “Will there be a war between countries A and B?” may itself affect whether the war happens.

If the model says there is a 1% chance of war, little changes. But if it says 90%, governments, markets, militaries, and the public may react: capital flees, troops mobilize, diplomatic trust collapses, and each side starts preparing for the other side’s preparation. The prediction helps make itself true.

The same feedback loop could apply to bank runs, market crashes, civil unrest, elections, and corporate failures.

At some point, the most accurate forecaster may become less like an observer and more like an actor with enormous power over the system it predicts.

A type like NonEmptyString is a weak defense against that, as a lazy dev can just pass a single space character or something similar.

Usually, you run the actual packaged dependency code at some point anyway, and usually with the same permissions as the install process.

So all of these setup scripts (good or bad) can just move their entrypoint from npm to wherever the `import` or `require` happens.

It seems to me that this is a small stumbling block at best, unless the whole ecosystem moves to a deno-like sandboxed environment. Maybe that is the plan?

SETI@Home is a very niche use case

and web browsing still happens by connecting to data centers and server farms, not by connecting to another laptop

Browsers are at least somewhat sandboxed

That is not how LLMs are typically used though in my experience

> Think of it like having a graphics card at home versus using a cloud gaming stream?

Latency seems to be much more important in that use case

1. in order to run LLMs, especially the best ones, you need complicated devices which are expensive

2. if you buy one for your personal use, you are probably not going to utilize it all the time and it will be idle a lot

It seems to me that it will always be more economical that the LLM-running devices are in a datacenter where it is easier to make sure they are always utilized

Most models don't have a 100% correct CLI usage and either hallucinate or use some deprecated patterns.

However `jj undo` and the jj architecture generally make it difficult for agents to screw something up in a way that cannot be recovered.

https://de.wikipedia.org/wiki/Norton_Partition_Magic#/media/...