Soyweiser @ Soyweiser @awful.systems

Posts

3

Comments

1,575

Joined

2 yr. ago

'we use LLMs for X in our security products' gets brought up a lot in the risky business podcast promotional parts basically, and it sometimes leaks into the other parts as well. That is basically the times I hear people speak somewhat positively about it. Where they use LLMs (or claim to use) for various things, some I thought were possible but iffy, some impossible, like having LLMs do massive amounts of organizational work. Sorry I can't recall the specifics. (I'm also behind atm).

Never heard people speak positively about it from the people I know, but they also know I'm not that positive about AI, so the likelyhood they just avoid the subject is non-zero.

E: Schneier is also not totally against the use of llms for example. https://www.schneier.com/blog/archives/2025/05/privacy-for-agentic-ai.html quite disappointed. (Also as with all security related blogs nowadays, dont read the comments, people have lost their minds, it always was iffy, but the last few years every security related blog that reaches some fame is filled with madmen).

It gets better btw, nobody mentioned this so far. But all this is over warnings. From what I can tell it still all compiles and works, the only references for the build failing seem to come from the devs, not the issue reporter.

E: I'm a bit tempted to send the guy a email to go 'I saw your blog and had a question, was it an error or did it stop compilation' but that would imho cross the line into harassment, esp as to be fair I think I should also divulge where I come from as an outsider which would not go over well with a guy in that kind of mindset (if I have him pegged correctly). The next blogpost would be about me personally.

Yeah took me a lot of effort and double-checking with translate. My German is bad.

I have not kept up but Scott did write an anti trump article again before the election. So we really cant blame them /s

I did some reading of his blog. Yep, prob for the best.

Ow god that thread. And what is it with 'law professionals' like this? I also recall a client in a project who had a law background who was quite a bit of pain to work with. (Also amazing that he doesn't get that getting a reaction where somebody tries out your very specific problem at all is already quite something, 25k open issues ffs).

E: Also seeing drama like this unfold a few times in the C:DDA development stuff (a long time ago), which prob was done by young kids/adults and not lawyers. My kneejerk reaction is to get rid of people like this from the project. They will just produce more and more drama, and will eventually burn valuable developers out. (E2: also really odd that despite saying he has a lot of exp talking to OSS devs, he thinks the normal remarks are all intended very hostile. "likely your toolchain setting it or your build script" and "I'll unsubscribe from this bug now" seem to me to be pretty normal reactions, one a first suggestion at what the problem potentially could be, and the other disclosing that he will not be working on the bug (holy shit the (non lawyer) guy being complained about here is prolific. ~100 contribs on average daily last week and an almost whole green year)). Also "I value such professional behavior very much" tags post with 'korruption'.

Another edit: Looked more at this guys blog and that are a lot of quite iffy opinions my man. (I noticed that the other post tagged 'korruption' talks about the how the AfD should be allowed to go against 'the rainbow flag' (I dont know the exact details of the incident), which while yes, legally ok, it still is a bit iffy). And then I scrolled more and saw this: "Deutschland braucht eine konservative Revolution! Warum wir uns ein Beispiel an den USA nehmen sollten" "Germany needs a conservative revolution, why we should follow the USA's example". He is a Musk/Trump/Venture Capitalist Manifesto true believer. Deregulate, stop the ideology build cars and go to space! The Bezos/Zuckerburg revolution. Common sense! "Musk, der Inbegriff des amerikanischen Unternehmergeistes" (If you allow me to react to this in Dutch: Lol). We need modern nuclear power, like how the USA does it (??). Deregulation, AI, humanitarian immigration that also only selects skilled workers, Freedom of speech which includes banning of "cancel culture", education reform, tax reform, stop crime, quantum computers, biotech, do more things online. We need to look forward, and change things, and thus a conservative revolution!

There is more stuff like: "Die temporäre Zusammenarbeit mit der AfD in einer Verfahrensfrage wird das Parteiensystem nicht nachhaltig beschädigen.", or https://seylaw.blogspot.com/2021/04/der-negerkuss-eine-suspeise-die-gemuter.html (If you don't speak German and want to listen to the weirdly racist drunking ramblings of a guy at the bar who is 'joking' throw it through google translate).

E: also forgot, lol at him going 'just run these two bash scripts I provided only takes 30 secs' like the devs need not first check of none of these is doing something malicious.

Yeah I used my political positioning in regards to the Dutch political people and the self identification of our political groups (the most rightwing parties after Wilders are self proclaimed liberals here after all (however talking of political ideology and what it implies is a thing that simply never happens in politics here, it is really strange, everybody acts like they have no real nameable ideology (the people not in power have of course))). Which colors my perception of these kind of things.

Cool thanks for doing the effort post.

My (wildly optimistic by sneerclubbing standards) expectations for “LLM agents” is that people figure out how to use them as a “creative” component in more conventional bots and AI approaches

This was my feeling a bit how it was used basically in security fields already, with a less focus on the conventional bots/ai. Where they use the LLMs for some things still. But hard to spread fact from PR, and some of the things they say they do seem to be like it isn't a great fit for LLMs, esp considering what I heard from people who are not in the hype train. (The example coming to mind is using LLMs to standardize some sort of reporting/test writing, while I heard from somebody I trust who has seen people try that and had it fail as it couldn't keep a consistent standard).

I now wonder how that compares to earlier non-LLM AI attempts to create a bot that can play games in general. Used to hear bits of that kind of research every now and then but LLM/genAI has sucked the air out of the room.

Tante on the hype term 'agentic AI'

Know it was a blog, so I meant my post as a bit of a joke, as this isn't any proof of a general FT hate. (but like all jokes, it isn't just a full joke, as I have noticed that a lot of financial media like FT (there is also a dutch tv show for example) randomly can be a lot more pro 'leftwing' financial stances (I mean the more left-liberal/socdem ones, not full blown fully automated gay luxury space communism) than you would expect (This is also a bit of a blindspot for a lot of more online leftwingers).

That whole 'haters just like incomplete information' makes no sense btw. Amazing to get that out of people basically going 'we hate beff for x,y,z' (which I assume happens, as I don't keep up with this beff stuff, I don't like these 'e/acc' people).

People might not be aware but places like FT seem to really dislike the techbro people

Ah so that is good news, worried it might be offline due to legal troubles, but it is technical troubles. Thanks for all your (and everybody else involved) service an all that.

it was written in latex

Of course it was. (I enjoy LaTeX myself but still, it is a type of person)

on how trying to make programming languages easier in some ways is doomed to fail

This is prob right, but the 'in some ways' part does a lot of work here. Think the issue is that some complexity can be removed without problem, and some absolutely cannot. And the problem of figuring out which is which is hard. (Which if you squint, seems to be similar to the chemistry stuff you describe here). With software it (as far as I can tell) is also quickly that bigger projects need bigger teams, and that adds a lot of communication problems, and as a non-stacking process you can't just add more programmers to make stuff go faster (compared to for example building a building, which can be sped up a lot more with just more workers) as these communication problems remain. From what I heard is that this, and the problem of maintaining software on a large scale is what Java was trying to fix. Which is why all programmers love Java. It is a language for enterprise scale projects. (On that note, which is also why a lot of reason people hate Java for the wrong reasons, a lot of the hated stuff makes sense if you recall it is made for enterprise scale projects/teams etc. It is an attempt to make those projects easier (lets leave it in the middle if that attempt worked or not (Do think it is amusing that Minecraft of all things was coded in Java by a single person (initially))).

Interesting our community seems to attract a few outspoken chemistry people. Not something I know much about, know somebody who does something with crystal chemistry machines, and when he technically talks about it I'm happy I understand about 30% :).

I'm sorry why did you make the robot so sexy?

looks up from recording my new mathematically speaking 'a podcast for the new thinking man' podcast

A phew, I was worried for a moment there.

E: Apologies, it is real I should have googled it. I know nothing of the podcast, I just tried to make a 'this is what a Rationalist/logic bro would name their podcast' joke. Ow god he even has an episode about conflict theory (but in contrast to Scotts post on conflict theory he actually talks about a historical mathematician so not the same thing, but that was a moment of double take). There is also an Adam Allred who is a 'masculinty speaker' or something, who is also into AI, Maga and everything else of course, but not sure if they are the same person (nope different people, turns out if you are called Adam Allred you are forced to make a podcast). But the math podcast Adam seems to be a good guy who is pro lgbt/BLM etc. (He did get his twitter account hacked which is now spamming people).

Unrelated to my recent posts on sciencefiction, and not sure if this is something I should ask here publically but it is the easiest place I could think off. But @dgerard@awful.systems is Rationalwiki dead or not?

That is a bit weird, as iirc the robots in star wars are not based on LLMs, the robots in SWs can think, and can be sentient beings but are often explicitly limited. (And according to Lucas this was somewhat intentional to show that people should be treated equally (if this was the initial intent is unclear as Lucas does change his mind a bit from time to time), the treatment of robots as slaves in SW is considered bad). What a misreading of the universe and the point. Also time flows the other way folks, LLMs didn't influence the creation of robots in SWs.

Also if the droids were LLMs, nobody would use them to plot hyperspace lanes past stars. Somebody could send a message about Star Engorgement and block hyperspace travel for weeks.

But yes, the backlash is going to be real.

E: ow god im really in a 'take science fiction too seriously' period. (more SW droids are not LLMs stuff)