ef2k

A significant part of the cost is that newer framework APIs are not known to the current agents. So they need to discover them and learn them on the fly. Eventually future models will have them trained into their weights and efficiency should increase. In the grand scheme of things, the cost might be negligible if it makes up for it in performance or quality.

I found that it's not just text. It's also design and overall user experience. If you have an eye for it, your brain picks up on it immediately with the little pills, busy landing pages, the use of serifs everywhere, bronze accents, and lots of dark mode.

Anythign that needs to be cleaned up or undone goes in ctx.effect. It returns the "inverse" (the cleanup function) when the plugin loads. Cordis then stores it and runs it when the plugin unloads. Take a look at 5.1.1 in the paper.

What's buried under the lede: this harness is using Cordis v4 (the paper that dropped today). Cordis has already been used for four years in a different project called Koishi that uses v3. Cordis itself is a way of hot loading and unloading plugins without restarting a running process. The cool part is that when it unloads it can revert any state and side effects it created, cleaning up its connections, memory allocations, registered handlers, etc. and it can also deactivate any dependencies it relied on without disturbing other plugins.

I don't know if it's a coincidence but this is the same thing an AI explained to me when prompted something like "what replaces SaaS now that AI can build most things on demand". That was like a year ago. I have no proof for this, but we might all just be getting similar responses from similar contextual bubbles and landing on the same conclusions.

to be fair, #2 exists because monorepos and submodules are somewhat antithetical concepts. A monorepo is supposed tobe the single source of truth for the codebase, while submodules are pointers to external repos with their own history. That alone will increase the source of churn for teams that are constantly merging.

> All my finance and payment domain expertise, all the debugging intuition and distributed system knowledge earned through hours of sweat and tears, is now promptable.

I think the author downplays how much of that knowledge is used on knowing what to zoom in on, what to prompt, or what to look for.

perfect time to drop a reminder: vendoring might seem annoying, but it's way safer than trusting a remote registry.

It's telling how rhetoric and conjecture are now normalized in company and business culture. When we were at the peak of remote work, companies were reporting record high revenues.

The issue was a compromised build pipeline that shipped a poisoned package.

But PSA: If something is critical to the business and you’re using npm, pin your dependencies. I’ve had this debate with other devs throughout the years and they usually point to the lockfile as assurance, but version ranges with a ^ mean that when the lockfile gets updated, you can pull in newer versions you didn’t explicitly choose.

If what you're building can put your company out of business it's worth the hassle.

A few years ago, intentionally fingerprinting or tracking your users without disclosure was spyware and unethical. Alas, here we are.

Anyway, what they're calling "spectroscopy", is a combination of extension probing and doing residue detection (looking for what extensions might leave behind in the DOM).

An ad blocker is not necessarily equipped to help since the script is embedded with the application code. Since they're targetting Chrome, switching browsers will help with the probing but not the detection part and you'll still be fingerprinted.

The only way forward is for browser vendors to offer a real privacy or incognito mode where sites are sandboxed by default. When the default profile is identical across millions of users there won't be anything unique to fingerprint.