sensarts

This is the key insight. A human's intent holds for the whole command. An agent's intent is rewritten every token. You can't auth them the same way.

This is the kind of post that makes me wish HN had bookmarks. The open-iscsi IQN slash issue alone was worth the read. Great work.

Simplicity always matters. Love the premise. Spreadsheets are the universal solvent until they become the universal problem. The domain renewals tracker is genuinely useful. Great work.

Great to see a survey like this. The GPU energy problem isn't going away. Real question for the authors: what about I/O overhead? A memristive array might look amazing on paper, but once you add DACs(dig.analog converter), ADCs(analog digital conv.), and a microcontroller to run the show, do the gains still hold? Does the paper count the peripherals or just the pretty core numbers?

I think Puter js would give you that capability, especially if you want a quick frontend demo with zero setup. But it is frontend only.

This is one of those tools that should exist soon or later. The fact that swapping mcp configs and auth methods for claude code requires a whole switcher tells me dx isn't there yet. Thanks for sharing.

This is a great example of how a simple, focused tool can make public data easily accessible. Good job.

When you have 5 workers + a judge all running in isolated VMs, what is a workflow for tracing a failure? Can I replay the event log locally, or inspect what each agent tried? And also, can agents share intermediate logs, results without going through the event system?

This is really cool. That's the good stuff. Did you notice any pattern in why models cluster? Shared training data or just similar architecture choices?

I think intelligence can be trained. In the school, I remember it was a slogan on the wall "Math is a brain's training" or something like that :) Removing laziness from your life, learning something and practicing it or implementing it somewhere make you more intelligent on that specific field.

Exactly, that's the worry. If two similar updates happen close in time, orthogonality forces them apart for no reason. I like your weight preference idea. Maybe just apply orthogonality when updates are actually different, not when they are nearly the same. Probably you should try that if not tried it already. :)

What's wild is that ClaudeCode used to feel like a smart pair programmer. Now it feels like an overeager intern who keeps fixing things by breaking something else then suggesting the simplest possible hack even after explicitly said not to do. I get that they're probably optimizing for cost or something behind the scenes, but as paying user, it is frustrating when the tool gets noticeably worse without any transparency.

The orthogonal LoRA constraint is interesting. Have you thought about whether orthogonality conflict with the timestamped training? If two temporally adjacent observations should produce similar LoRA updates, orthogonality would actively push them apart. Maybe you want similarity for recency, orthogonality only for distinct episode types?

Is this inspecting certificates in real-time or analyzing stored traffic? And does it flag things like expired certs, weak ciphers, or pinned certificates that failed validation?

wow, more than 10K airbuses just over the NA. Looks like airplanes are the major source of air pollution.