14u2c

More likely we'll just see another shakeup of the layers of abstraction that the average company operates at. Why would you hire specialized people when AWS can do so for you?

You're definitely not wrong. It's the new code smell.

Yes, exactly. In other words, they were introspective.

> Great men of history had little to no introspection.

These people are completely detached from reality. How do they think we know the thoughts of these so called great men?

> No one is stopping them.

Except for the person refusing to grade it, of course.

I should have done so, that’s unfortunate. Doesn’t bode well for their Java stack.

Jetbrains software is not web-based.

> You have to figure out a new transactional structure that can’t have your tuition made irrelevant by a $20 subscription.

Maybe a problem in the long term, but irrelevant here. These students paid (a lot of) money for this service. The provider of the service needs to stop complaining and do their job (grade the homework). If even a single student did the work themselves, anything otherwise would be a disservice.

> even that's fine because output tokens cost ~5x input tokens.

That's assuming you come in under a certain threshold of revisions, and excludes the tokens generated for all the exchanges. The later can be be significant, especially when there are significant issues to correct.

You are correct, at least based on my experimentation. This subagent workflow really only delivers meaningful results when working across model families (ex: Opus asking Astra for a review) or when working across conceptual hardware boundaries (cloud offloading to local inference, etc). The planner / overseer model ends up reading all the output of the coder model anyway, at least if you want quality output, so there's not much benefit otherwise.

> A user-space daemon can saturate almost any network link using a single CPU core.

Certainly not a guarantee at today's link speeds. We're increasingly leaning into RDMA and other forms of offloading for a reason.

This comment has major 2018 vibes. Most enterprises I know have have migrated back to mainline Java. We'll see if this changes again LLM era, now that prioritizing the talent pool is less important and the friction of migrating stacks is much reduced.

Maybe if you are considering crime committed in pursuit of profit, but unfortunately there can be many other aims.