When I scroll up in nvim, it will keep the editor frame in place (that's the top bar and bottom bar showing things like open buffers, git status, the scratch buffer or whatever it's called), but the file contents will scroll by because nvim at that point has exclusive ownership of the entire screen and can do anything with it, including repainting parts of it in response to motions or a mouse scrolling (if your terminal supports emitting mouse events).

This is in contrast to the `rmcup` "normal" terminal mode where it will scroll back in the terminal's history.

The best analogue I have for that last one is to use tmux with nvim open, and have a tmux visual selection going. You can scroll up and out of nvim, and keep scrolling to whatever was executed before neovim, and when you get out of tmux visual mode it'll snap back down to the bottom of your scrollback buffer, nvim (nominally) taking up the entire pane like nothing happened; but we can probably agree that outside of a few narrow use cases, this isn't a very desirable way to manage scrolling in a terminal.

With the usual terminal mode, that history can outlive the Claude application, and considering many people keep their terminals running for days or sometimes even weeks at a time, that means having the convo in your scrollback buffer for a while.

The theory states that Anthropic avoids using the alternate screen (which gives consuming applications access to a clear buffer with no shell prompt that they can do what they want with and drop at their leisure) because the alternate screen has no scrollback buffer.

So for example, terminal-based editors -- neovim, emacs, nano -- all use the alternate screen because not fighting for ownership of the screen with the shell is a clear benefit over having scrollback.

The calculus is different when you have an LLM that you have a conversational history with, and while you can't bolt scrollback onto the alternate screen (easily), you can kinda bolt an alternate screen-like behaviour onto a regular terminal screen.

I don't personally use LLMs if I can avoid it, so I don't know how janky this thing is, really, but having had to recently deal with ANSI terminal alternate screen bullshit, I think this explanation's plausible.

    > FastMCP is the standard framework for building MCP applications

In an era where technology exists that can lend the appearance of legitimacy to just about anyone, that kind of statement needs to be qualified.

A few friends and I have a small handful of self-hosted services that we all run on a VPN between our places with stuff like a recipe sharing app, etc., but the number of people with access to that is single digits.

In terms of "hosting anything," I still have my own homelab, and my self-hosting will be limited to this sort of stuff for the foreseeable. A cluster of limited-scope apps that helps me and a handful of friends keep in touch after moving out of our hometown, beyond just chatting in Signal groups.

I won't be putting up my own public website (or portfolio, or whatever; be it hosted on my own infra or not).

I've picked up hobbies since then that lend themselves well to sharing online, but right now, with the amount of LLM-related scraping happening, I have no intention at all of hosting anything I've made by hand, be it code, photos, recipes, etc.

These bots make their own ground rules -- "just put up this special robots.txt thing" -- and then ignore them, requiring tools like Anubis to be created, maintained, and kept constantly up to date, and for what? So the plagiarism machine can plagiarise more? Copyright courts have apparently just checked out on this, so I'm not at all confident they'll do anything about this.

To be clear: my ego isn't so huge that I want my name plastered everywhere, on everything I've ever touched. I believe strongly in the principle of not getting your shit scraped online and then churned out into something that these LLM companies can (eventually? presumably?) profit off of (or at least turn into investment, in the short-term). These companies have done terrible things to the state of public discourse and I do what I can to avoid feeding the beast.

    > Based on the fact user scripts are globally disabled now I'm guessing this was a vector.

Browsers still allow for user scripts via tools like TamperMonkey and GreaseMonkey, and that's not enforceable (and arguably, not even trivially visible) to sites, including Wikipedia.

As I say that out loud, I figure there's a separate ecosystem of Wikipedia-specific user scripts, but arguably the same problem exists.

Neural networks, computer vision, sentiment analysis, all of these and more have provided an unspeakable amount of value over the years.

There's a reason most people in well-banked countries use plastic over cash these days, convenience and consumer protection are it; and even with cash, the barrier to entry for fraud is higher since at some point that cash will come into contact with the banking world, and will have to be accounted for. If your cash flow is suspicious enough, it'll be audited.

- They provide no meaningful consumer protections (since this necessarily requires an authority, which blockchains may not have)

- They don't protect at all against meatspace vulnerabilities like scams and other deception-based attacks, which are by far the more common issue in banking. This is exacerbated by the lack of consumer protections.

(To be clear: don't read my comment as being in support of PayPal. They have abused user trust for a while, and I haven't had an account there in over a year -- fuck 'em.)

It also doesn't get in the way of the US's (already extraordinarily loose) firearms sale and acquisition doctrine which regularly costs innocent lives with the pretext of resisting a tyrannical government (which, it appears, people aren't actually that interested in doing now that one's in power).

So, yes, technically, it's de-facto advisory to publish this information, but assigning "advisory" as a severity tag here is questionable.

I look forward to Bartosz's articles because they're rock-solid sources of information and the visualizations are both easy-to-understand and surprisingly light on performance. It's all shockingly digestible.

Honestly, as popular science writing goes, this is art as far as I'm concerned, and art is best when it comes from a place of passion and conviction, something AI will never be able to reproduce.

It's like when I learned that many paper recycling programs end up combining paper with regular garbage, or finding out that plastic recycling is comically ineffective in its outcomes.