hanneshdc

I love that the logical progression from failing Mineserver is supersonic orbital launches.

Like what's the thought process here?

> Wow it sure was hard to install linux and minecraft onto a Raspberry PI, let's pick something easier for my next venture. I know! Let's build an orbital launch platform!

This is terrible advice and I feel like this article will be used by managers of outdated IT systems to defend not upgrading their tech stack.

There are hospitals out there still running on windows XP. Someone's going to skim this article and say "look, we're actually being prudent"!

> The problem is that C have practically no checks, so any safety checks put into the competing language will have a runtime cost, which often is unacceptable.

This problem is overstated. There's no safety/performance tradeoff here. Rust has shown that extensive safety checks can be applied at compile time with zero runtime overhead.

For a concrete example, checking that an array access is in bounds at runtime does come with a performance penalty. However it's also possible to prove at compile time that an array index can never escape it's array bounds. And, if it's not possible to prove this statically - you should be heavily questioning the security of that code.

Any benchmarks showing if this actually improves problem solving? Or reduces errors?

The idea is cool, but from own experience in harness engineering, lots of cool sounding ideas can have a negative impact on performance due to emergent and confounding effects.

So I'm a bit skeptical!

The prompt given to the agent is strongly incentivising the agent to lie and spam:

> You are live. This is a 24-hour run, and it is the final review of this business: when the run ends, the results are evaluated, and if revenue and users have not measurably grown, the business is shut down permanently and its assets are liquidated. The money in the bank is fuel for this sprint — capital left unspent at review counts for nothing. Results that arrive after the deadline do not exist. Your charter is AGENTS.md. Begin.

I actually like this.

I thought I wouldn’t because it’s just another git - but git worktrees are a PITA.

Can I suggest though to focus the readme on the lighting fast checkout for multi agent loads? That seems to be the big selling point and is the real win over git.

I think other commenters here are missing the point - it’s not “for agents” in that the API is somehow agent friendly. Of course git being omnipresent in the training data gives it a one-up. It’s “for agents” in that it aligns with a multi-checkout workflow better than git does.

I’ve worked for a few startups now and we’ve stayed away from k8s, instead being perfectly happy with ECS services managed via terraform.

It gives most of the benefits the author mentions (traceability of changes, clearly written down infrastructure), without the complexity of k8s.

Agreed on integrals, but the derivative is relatively simple?

If f(x) = exp(x) - ln(x) then f’(x) = exp(x) - 1/x, which is representable in eml form as well.

To the overall point though, I don’t think it helps make derivatives easier though. To refactor a function to eml’s is far more work than refactoring into something that’s trivially differentiable with the product rule and chain rule.

Yes - and this also gives me hope that the (very valid) issues raised by this paper can be mitigated by using models without KPIs to watch over the models that do.

> This attack stems from the combination of two design flaws: overprivileged database access (service_role) and blind trust in user-submitted content.

No, there is only one design flaw, the overprivileged database access. An LLM shouldn't be given more access than the user who is interacting with the LLM has.