LLMs in medicine. What could go wrong?

There’s nothing fundamentally wrong with LLMs.

Disagree.

LLMs should die in a fire.

I can’t imagine it’s going to be very long before Elon’s hostile attitude toward basic safety results in a high-profile catastrophy involving human deaths under the ospices of Space X.

Honestly, this isn’t much of a hypothetical for me. At work, my choices are Windows, Mac, or Ubuntu. I’m quite happy with Ubuntu, though I’ve switched away from the default desktop environment to i3.

I use Arch (BTW) on my personal systems. And Ubuntu isn’t as bad as I worried it would be.

My main gripe is snaps. Firefox is practically unusable as a snap. And my employer forbids installing any software (save for a select list of exceptions) not via the officially-supported Ubuntu way of doing things. Chrome is available without snap, so I use it on my work machine. Which annoys me, but if I’m less efficient in my job as a result, it’s their own fault.

And how much are you asking for in research funding?

I was really excited for CJDNS (not to be confused with “Domain Name System”) at one time. It’s a mesh networking protocol. And they’d established an “Internet 2” (as in, a sequel to “The Internet”) based on the CJDNS protocol called “Hyperborea.”

I haven’t heard anything about CJDNS in a good while now. I wouldn’t be surprised if there were other efforts looking to do something roughly the same, but I’m not up to date on anything more recent.

Llama is not Open Source

I kindof hate the slogan “they go low, we go high” (from Hillary’s campaign.)

But this is an example of the “good” side of that slogan. The political left(-of-what-passes-for-center-in-the-U.S.-now-a-days) isn’t given to publicly calling for assassinations of the opposition party. It’s not even given (and, yes, there are exceptions) to calling privately for assassinations of the opposition. And that’s a good thing.

It means the left(-of-U.S.-center) hasn’t turned into the fascist-dictatorship-trying-to-happen that the right has. It’s not the left(-of-U.S.-center) calling for civil war and pandering to creeps who chant “blood and soil” while carrying tiki torches around the capital.

The day left(-of-U.S.-center) news sources delight in assassinations even of opposition as dangerously unhinged and power hungry as Trump because that sentiment started with snide remarks like yours is the day we have to worry that maybe the Democrats are sliding into their own brand of fascism.

Don’t get me wrong. I’m for radical support of LGBT rights, womens’ autonomy in matters of personal health, universal free healthcare, and most other “liberal” causes. (I also identify as well left and libertarian-ward of the Democratic party and would love to see “to each according to need” be our modus operandi. I’m also for direct action.) I don’t fault the Democrats for being “too radical” by a long shot. (More likely, the Democrats will continue to be far too willing to let the Republicans control the narrative and cheat their way to political power. And that’s the bad side of “they go low, we go high”) And I don’t believe it’s very likely that the Democrats will slide into widespread advocacy for political violence like the Republicans have much more so already.

But taking delight in assassination attempts and wishing they’d been successful – even those directed at Cheeto-flavored Hitler himself – isn’t helpful.

All that said, I get it. I’m pissed at the U.S.'s descent toward fascism, too. But wishing him assassinated isn’t going to change anything for the better.

In this thread: Cryptobros downvoting every realistic take on cryptocurrency for being “bearish” and “FUD”.

I cannot think of any other methods

Exactly. What you’re describing isn’t “AI.” It’s “magic.” And “AI” can’t do what OP wants either.

No “AI” solution we have any reason to expect we’ll be able to create in anything approaching the foreseeable future is going to be able to do anything remotely like this without ridiculous amounts of false positives and/or false negatives.

By false positives in this case, I mean things like not coming back from the cool little slideshows until a minute past the end of the commercial break or obscuring important details of the show having falsely “concluded” that it’s a logo or some such.

And I would have assumed “without a lot of false positives” would have gone without saying. If OP is comfortable with lots of non-ad content blocked/obscured along with the ads, then I’ve got a 100% guaranteed zero-false-negatives solution that’ll fit OP’s requirements without involving a speck of “AI” anywhere that OP can implement right now: turn the TV off.

A pet pieve of mine is people randomly sticking the term “AI” into a description of some particular tech solution.

You want ad blocking. (Which is based.) But you don’t want “AI”. If this can be done in a way that doesn’t qualify as “AI”, that would satisfy you, yes?

And using the term “AI” that way makes it clear you haven’t really thought through what you really even want in that feature. (Not that there’s anything particularly wrong with that, especially in a showerthoughts community, but it’s still kindof a “slaps me in the face” kind of thing.)

And the term “AI” is so imprecise anyway.

And particular kinds of “AI” are such a bubble right now. And that’s why everybody is sticking the word “AI” into random contexts for no fucking reason. But it’s also just a gimmick at best and a huge scam at worst.

And “AI” is inevitably bad about false positives and such.

I’d really rather see the word “magic” than “AI” in this context. Because at least that admits that this is an idle wish and not something you think actual real-world adult humans should be seeking venture capital to attempt.

I’m sorry for taking this out on you specifically. You’re definitely not the first person I’ve seen do this.

I write Java for a paycheck, but I really hate it.

It feels like everything is layers and layers of overengineered cruft, each added to the precarious tower for something extremely minor. But every subsequent card in the house of cards makes it more precarious. “But look, I don’t have to write accessors.” “But look, I eliminated the need for the web.xml file.” “But look, I don’t have to understand SQL now.” But look, the codebase depends on a shit-ton of completely opaque Automagic™ that you have no hope of understanding the moment something goes wrong – which it will if you even think of changing your Java version. And since it’s practically impossible to understand what’s going on under-the-hood of whichever dependency is fubar’d this week, you have to resort to a mixture of trial-and-error and copy-pasting shit (that you also don’t understand) from StackOverflow and praying to Cthulhu something works – which is also trial-and-error because Java questions in particular have tons of just straight up wrong answers.

To be fair, I’m the guy on my team who people come to when they run into those sorts of “I bumped up one subminor version of Mockito to fix a bug that was preventing my unit test from working but now literally half of our unit tests won’t build” or “I added the war plugin to the build.gradle and now SwaggerUI is broken.” So maybe I see more than my fair share of “well shit, I guess I’ll just spend the next three hours hunting down which magical combination of Jar version numbers will fix things” kind of problems. But damn. This shit didn’t ever happen back when I was doing Python for a paycheck.

I don’t use Java if I don’t have to. If I have to use Java, I prefer to just use Servlets (mostly I do web development) and absolutely as few dependencies as I can possibly get away with. Fewer moving parts mean less that can break.

This absolutely sent me.

“work”

The other three quarters are just scared that Elom will sue them if they cut advertising.

(Not really. I suspect many of the other 75% just aren’t willing to admit they’re planning to loosen ties with Twitter (I will not call it “X”) just yet.)

The A* algorithm doesn’t have anything to do with machine learning either, but the first time I ever learned about it was in a computer science class in college called something like “Introduction To Artificial Intelligence”.

But it’s very much the case that the term “AI” has a very different meaning now-a-days during this cringy bubble than it did back in 2004 or 2005 or whenever that was.

Today “AI” is basically synonymous with “BS”. Lol.

AI is quite fit for the task of understanding what might be the purpose of code

Disagree.

I don’t know how some non-AI tool could be better for such task.

ClamAV has been filling a somewhat similar use case for a long time, and I don’t think I’ve ever heard anyone call it “AI”.

I guess bayesian filters like email providers use to filter spam could be considered “AI” (though old-school AI, not the kind of stuff that’s such a bubble now) and may possibly be applicable to your use case.

I don’t think “AI” is going to add anything (positive) to such a use case. And if you remove “AI” as a requirement, you’ll probably get more promising candidates than if you restrict yourself to “AI” (whatever that means) solutions.

Sweet! Now let’s all go commence scowling at Redis until they do the same.