@reflex

No, this has not consistently been shown to be true. You might think so because Mastodon - really - is a little weird bubble here.

I'll take my own profession as an example. I'm a very senior software dev turned cybersec. Walk outside of Mastodon and there's no question whatsoever that LLMs are _useful_ both in software dev and as both reverse engineering and red teaming agents.

Here's Linux kernel devs:

https://www.theregister.com/2026/03/26/greg_kroahhartman_ai_kernel/

Bagder of cURL fame says the same thing.

If you really believe differently I'm sorry but you're ... out of touch.

@rootwyrm @davidgerard

@troed @rootwyrm @davidgerard Dudebro, my career is verifiable, I spent 11 years off and on at Microsoft in several roles including Windows kernel engineer, I spent 8.5 years at Amazon including being the architect of the "AI" driven network we used for the AmazonGo stores. I have tons of former colleagues in both places.

They do NOT agree with your experience and are miserable having to 'prove' AI's usefulness to execs for the quarterly reports.

It's shit. The emperor has no clothes.

@reflex

I didn't source _my_ experience now, did I? What were my sources again?

Here's Bagder from today: https://mastodon.social/@bagder/116354106408236089

Or let's take a well known writer just now:

https://mastodon.social/@harrymccracken/116358878961334760

@rootwyrm @davidgerard

@troed @reflex @davidgerard uhhuh, love the intentional deletion of context and refusal to discuss or accept any facts. While trying to argue with some of the leading experts in the world.

So what, are you a sock puppet for some slop peddler child? Or just a worthless slop peddler desperate to preserve your self-esteem as you get completely and utterly destroyed on the Internet?

@reflex @davidgerard gotta love slop gobbling clowns that insist they are entitled to sealioning with multiple experts.

Including one of the guys who helps journalists unwind all the fun, fabricated, fantastical financials, hardware hilarity, infrastructure idiocy, and data center dishonesty.
Who also happens to have done stuff with real NN since the 90's.

@rootwyrm

Not that I believe you to be debating seriously, but ok:

I am one of those persons who wrote back-propagating neural networks in the early nineties. I'm also an LLM-for-coding skeptic that has changed his mind after actually having put my convictions to the test and used it in various situations. Oh, and yeah, I am that expert in *hands waving* shitloads of stuff. You're not argument-from-authority-winning here.

@reflex

I select which assignments I take on. I have no need to cheer lead anything. As I wrote to you, but I'm not sure you understood, I'm the person saying _no_ to AI development where I consult right now (since it's in a sector that needs secure coding).

I also use Mistral LLMs, since they do care about what data they train on.

@davidgerard

@troed @rootwyrm @davidgerard In other words, yes you should know better, and likely do know better, but it's more profitable to pretend the sky is purple. If you are as experienced as you claim and you are busy arguing against the fact that AI generated code is a shit sandwich, you are literally lying for some personal motive.

There is no 'debate' to be had here. It's shit code. If you are going to claim it's not when we can all see Claude Code, you are either incompetent or lying.

@reflex

Or I'm simply more competent than you are, and can accept the fact that LLMs are useful for a lot of tasks even though they're not suitable for some.

I laugh at everyone producing public SaaS written with LLMs - since I'll be able to charge lots of money fixing all those security holes.

The two mods for Hytale and the local meshtastic network planner I've written with LLM aids are doing just fine though.

I guess in your world everything is always either/or.

@rootwyrm @davidgerard

@reflex

What did I just write about those that use LLMs to put SaaS into public production ... ?

That doesn't mean they're useless. Here - watch me hack an IoT device using very low level reverse engineering. Then recall what I said about being impressed with the RE an LLM did on a fully proprietary binary.

I know my stuff. I'm saying they're useful. That doesn't mean that they're 100x coders taking over the world.

https://video.troed.se/w/kfbeBKcDuZt2KcyMx2kfsq

@rootwyrm @davidgerard

@reflex

see previous reply

@troed In other words you won't answer. You know how the answer would reflect on you.

I think we are done here. Good luck with your slop.

@reflex No, I'm saying I have answered and I'm waiting for you to acquire knowledge before repeating the same things over and over.

The difference between us is that I put my beliefs to the test. You haven't.

Of course I maintain the three apps I've made with the aid of LLMs and published. They're even open source - anyone's free to have a laugh. They do the job, and they're not security critical.

What other areas in society do you believe benefit from you voicing your uninformed opinions on?

@troed @reflex do you have no qualms about supporting a fraudulent product? whether or not the stochastic LLM device happens to generate output that you find personally useful is irrelevant to the larger issue of fraud: LLMs are sold as "intelligent" and "artificial intelligence" even though they work by a non-thinking mechanism, and intentionally lack the ability to discriminate between truthful and false information—one of the key features of genuinely thinking beings. They freely confabulation truth with garbage, yet they're marketed and sold as if they were smarter than human beings.

Doesn't that fraud bother you? Why are you supporting this scam?

@mxchara I'm not really sure what to focus on in this reply :D I spent a considerable amount of time studying the topic of consciousness a bit more than a decade ago, and in general us humans assume way too much about our own capabilities.

But to your point about LLMs - I haven't seen Mistral AI claiming anything of the sorts, and it's their models I'm using. They're in no way better developers than I am, but they're very quick at a lot of tasks which means I have to write less "boilerplate" and can instead focus on the important parts of the applications.

In some ways it's like using a high level language instead of writing everything in assembly code.

@reflex

@mxchara

As I said, this is a topic I have studied extensively. AFAIK science is not yet at the point where we can define neither intelligence nor what gives rise to the feeling that "we" exist as beings capable of introspection.

Back in the 50's we thought we would have natural conversations with computers in the next few years, while we believed they would never be able to beat us at chess. As soon as we solved how to do chess we stopped considering it as something that needed high levels of intelligence - but natural language took way longer.

My go-to regarding consciousness has been Blackmore's books on the subject. I'm thus not a dualist, and if that's the road we're going down we'll likely never be able to agree on even simple definitions ;)

@reflex

@troed @reflex If you've studied it extensively then you must surely be able to communicate your understanding, with a simple working definition of what "intelligence" is. I think that's important when discussing the activities of profit-seeking entities looking specifically to monetize the concept of "intelligence". I say that LLMs are not actually intended to be intelligent—intelligent behavior would interfere with the desired use-cases of the LLM devices. They are intended to be repetitive, predictable in their responses, devoid of true creativity and ability to generalize in favor of the ability to regurgitate massive amounts of undigested training-text as quickly as possible.

Surely these devices aren't intelligent. I see no evidence that they are meant to behave intelligently. You seem to think otherwise. Hence it would be nice to know: how do you define intelligence?

@mxchara

I wrote about their emergent capabilities here: https://blog.troed.se/posts/emergent_parrots/

Since science is unable to even come close to explaining consciousness I don't really understand why you think it's simple to define "intelligence".

How much have you studied the topic yourself, and do you adhere to monism or dualism?

@reflex

@mxchara I'm trying to have a high level discussion on concepts relevant to the subject.

Do you consider yourself a conscious being?

Science is unable to answer why that is. Regarding "basic and straightforward intellectual questions".

(That "outside material" is written by me - what's the difference if I write it here or not?)

@troed Yes I do. Now let me point out one of the consequences of that self-awareness: I am capable of discerning whether a statement is likely to be truthful or fallacious. This is a capability which LLMs are deliberately constructed to LACK, I remind you.

Do you quarrel with that characterization?

@mxchara Let's quote Wikipedia:

"There is controversy over how to define intelligence. Scholars describe its constituent abilities in various ways, and differ in the degree to which they conceive of intelligence as quantifiable."

Underneath follows a table with some definitions from some well known people in the area. They're very far from similar, and some would apply to today's computer programs and some would not.

Why do you claim this to be "simple"?

https://en.wikipedia.org/wiki/Intelligence#Definitions

@troed "Intelligence is complicated, therefore it's all right to pretend that LLMs are intelligent." Are you going to address the specific point about whether you believe in the marketing claims of the generative-AI firms?

I say that they're committing fraud. You're stuck on blithering about your own inability to communicate your understanding of a common English word.

Do you habitually use words without having a clear notion of what they mean?

@mxchara I just showed you that "intelligence" isn't well defined in science.

It seems you live your life by your gut feelings rather than basing your views in the current scientific consensus.

As it says in my bio: 'I discuss based in facts and actively despise "but my gut feeling .." arguments.'

I have not seen a single statement from Mistral AI that would constitute fraud.

@mxchara As I said, I debate using well sourced facts and I'm totally uninterested in whatever fantasies you come up with yourself to solve your cognitive dissonance.

(No, your guesses are not correct)

@mxchara All your guesses and fantasies are wrong.

You don't base your arguments in scientific facts. Thus they are useless.

@troed So, you don't profit from AI (real or fraudulent) in any way? Really?

Do you have a portfolio of tech investments? Yes or no answer, please. Surely you can manage a yes/no answer.

@mxchara Correct. I own no tech stocks and if you had read other posts in this thread you would know that I'm currently working at an assignment where I'm arguing _against_ the use of LLMs for coding.

You might want to ponder why you assume so much.

@troed I still don't believe that you have NO tech investments. That's very unlikely in the modern business climate, in which professionals (whether or not they're in computing) are encouraged to invest in the market, and "grow their wealth".

I think you're lying about that, for emotional reasons. You seem apt to lose your cool.

@mxchara "I still don't believe that you have NO tech investments"

"I think you're lying about that"

That's one way of handling your cognitive dissonance, of course. Rejecting facts than don't agree with your gut feelings.

I'm Swedish. We don't do the US "grow their wealth" stuff in general.

@mxchara Funnily enough, we now know you're not.

"I am capable of discerning whether a statement is likely to be truthful or fallacious."

@troed Insulting me isn't an argument. Do please stop wandering off topic. I asked you to supply your own working definition of a word, "intelligence". In reply you refused to answer directly, pointed me to irrelevant outside information, and finally you've lost your temper and turned into a very hostile witness indeed.

Is there some reason you can't give me an approximate definition of a word you use regularly? That's practically a schoolbook exercise.