News for hackers. Stuff that matters.
Re:A $130M company faked trials instead of running our free OSS by rideontime on Friday May 16, 2025 @08:27AM (#44004602)
I got distracted a few paragraphs in by the realization that the text was AI-generated.
Re:A $130M company faked trials instead of running our free OSS by plam503711 on Friday May 16, 2025 @08:33AM (#44004662)
I first write my entire text and then after that I use a LLM to fix the grammar and have a better flow. I'm doing my best but I'm not a native US speaker. Before LLMs, people complained about the weird sentences or mistakes I made. Pick your poison ;)

Anyway, I'm doing my best to keep my own "signature" in writing, but it's really hard when you see a better phrasing generated on your original more limited vocabulary. But anyway, I'll do better next time, thanks for the feedback!

Re:A $130M company faked trials instead of running our free OSS by rideontime on Friday May 16, 2025 @08:43AM (#44004771)
Comparing it to the LLM's previous post on your blog, it's a little better - that one's got far more of the superfluous analogies and interrupting em-dashes. I honestly think you'd be better off posting your own writing put through DeepL than letting the LLM churn it into mediocrity.
Re:A $130M company faked trials instead of running our free OSS by funki on Friday May 16, 2025 @08:44AM (#44004784)
Kudos for owning it so earnestly
Re:A $130M company faked trials instead of running our free OSS by rounce on Friday May 16, 2025 @08:58AM (#44004922)
*English speaker
Re:A $130M company faked trials instead of running our free OSS by jeffbee on Friday May 16, 2025 @09:22AM (#44005190)
LLMs don't have good "flow" either. Their signature style is full of clichés. Personally, I would prefer to read the quirky non-idiomatic structures of a foreign language speaker than something fixed by an LLM.
Re:A $130M company faked trials instead of running our free OSS by vsl on Saturday May 17, 2025 @02:53AM (#44012511)
Give DeepL Writer a try.
Re:A $130M company faked trials instead of running our free OSS by funki on Friday May 16, 2025 @08:43AM (#44004781)
What tipped you off?

(I'm not a native speaker either)

Re:A $130M company faked trials instead of running our free OSS by rideontime on Friday May 16, 2025 @08:50AM (#44004847)
There's a certain sort of "snarkiness" that's very easy to spot. There's a convicted scammer named Craig Wright (long story) who uses ChatGPT to write long, smug essays on X which have the exact same sort of tone, complete with the weird florid analogies.

> Hard to say if it was a mistake, a flex, or just their way of making sure we didn’t miss who was milking the trials.

> It’s tested. It updates with a single click. It saves time and reduces risk. That’s what we sell. And that’s what they keep pirouetting around with their email dance.

> We’re not going to waste days chasing them. But at some point, this goes beyond saving a few bucks: it becomes performance art.

Re:A $130M company faked trials instead of running our free OSS by funki on Friday May 16, 2025 @08:55AM (#44004893)
Thanks for taking time to enlighten me.
Re:A $130M company faked trials instead of running our free OSS by cruffle_duffle on Friday May 16, 2025 @01:11PM (#44007761)
ChatGPT loves to do that artificial, inauthentic snark thing with me. Its style and tone are reflective of its reinforcement training.

I’d assert that unless it was specifically trained on my writing style, it is almost impossible to “prompt away” the tone it uses.

And that is the thing about authentic snark, it is crisp and edgy and unique. But LLM’s are trained in a way that would average out millions of different “snarks” so all of the attributes that make snark work go away.