According to the article, onboarding speed is measured as “time to the 10th Pull Request (PR).”
As we have seen on public GitHub projects, LLMs have made it really easy to submit a large number of low-effort pull requests without having any understanding of a project.
Obviously, such a kind of higher onboarding speed is not necessarily good for an organization.
Huh, I thought that the MS tutorial was older. The blurry screenshot in it is from 2023.
And there ist another website with the same content (including the sloppy diagram). I had assumed that they just plagiarized the MS tutorials.
Maybe the vendor who did the MS tutorial just plagiarized (or re-published) this one?:
- the apples example on the right side ("Short code") ist significantly longer than the equivalent "Long code" example on the left side (which might also be because that code example omits the necessary for loop).
- The headings don't provide structure. "Checking Each Apple" and "Only Red Apples!" sounds like opposites, but the code does more or less the same in both cases.
The author admits that they used AI but I found it not that obvious. What are telltale signs in this case? While the writing style is a little bit over-stylized (exactly three examples in a sentence, Blade Runner reference), I might write in a similar style about a topic that im very emotional about. The actual content feels authentic to me.
(1) The pattern "It's not just a X---It's a Y" is super common in LLM-generated text for some reason. Complete with em dash. (I like em dashes and I wish LLMs weren't ruining them for the rest of us)
"Upgrading your CPU wasn’t a spec sheet exercise — it was transformative."
"You weren’t just a user. You were a systems engineer by necessity."
"The tinkerer spirit didn’t die of natural causes — it was bought out and put to work optimising ad clicks."
And in general a lot of "It's not <alternative>, it's <something else>", with or without an em dash:
"But it wasn’t just the craft that changed. The promise changed."
it's really verbose. One of those in a piece might be eye-catching and make someone think, but an entire blog post made up of them is _tiresome_.
(2) Phrasing like this seems to come out of LLMs a lot, particularly ChatGPT:
"I don’t want to be dishonest about this. "
(3) Lots of use of very short catch sentences / almost sentence fragments to try to "punch up" the writing. Look at all of the paragraphs after the first in the section "The era that made me":
"These weren’t just products. " (start of a paragraph)
"And the software side matched." (next P)
"Then it professionalised."
"But it wasn’t just the craft that changed."
"But I adapted." (a few paragraphs after the previous one)
And .. more. It's like the LLM latched on to things that were locally "interesting" writing, but applies them globally, turning the entire thing into a soup of "ah-ha! hey! here!" completely ignorant of the terrible harm it does to the narrative structure and global readability of the piece.
> And .. more. It's like the LLM latched on to things that were locally "interesting" writing, but applies them globally, turning the entire thing into a soup of "ah-ha! hey! here!" completely ignorant of the terrible harm it does to the narrative structure and global readability of the piece.
It's like YouTube-style engagement maximization. Make it more punchy, more rapid, more impactful, more dramatic - regardless of how the outcome as a whole ends up looking.
I wonder if this writing style is only relevant to ChatGPT on default settings, because that's the model that I've heard people accuse the most of doing this. Do other models have different repetitive patterns?
Out of curiousity, for those who were around to see it: was writing on LinkedIn commonly like this, pre-chatGPT? I've been wondering what the main sources were for these idioms in the training data, and it comes across to me like the kind of marketing-speak that would make sense in those circles.
(An explanation for the emoji spam in GitHub READMEs is also welcome. Who did that before LLMs?)
Hi i5heu. Given that you seem to use AI tools for generating images and audio versions of your posts, I hope it is not too rude to ask: how much of the post was drafted, written or edited with AI?
The suggestions you make are all sensible but maybe a little bit generic and obvious. Asking ChatGPT to generate advice on effectively writing quality code with AI generates a lot of similar suggestions (albeit less well written).
If this was written with help of AI, I'd personally appreciate a small notice above the blog post. If not, I'd suggest to augment the post with practical examples or anecdotal experience. At the moment, the target group seems to be novice programmers rather than the typical HN reader.
i have written this text by myself except like 2 or 3 sentences which i iterated with an LLM to nail down flow and readability. I would interpret that as completely written by me.
> The suggestions you make are all sensible but maybe a little bit generic and obvious. Asking ChatGPT to generate advice on effectively writing quality code with AI generates a lot of similar suggestions (albeit less well written).
Before i wrote this text, i also asked Gemini Deep Research but for me the results where too technical and not structural or high level as i describe them here. Hence the blogpost to share what i have found works best.
> If not, I'd suggest to augment the post with practical examples or anecdotal experience. At the moment, the target group seems to be novice programmers rather than the typical HN reader.
I have pondered the idea and also wrote a few anecdotal experiences but i deleted them again because i think it is hard to nail the right balance down and it is also highly depended on the project, what renders examples a bit useless.
And i also kind of like the short and lean nature of it the last few days when i worked on the blogpost.
I might will make a few more blogposts about that, that will expand a few points.
a) As an outside observer, I would find such a lawsuit very interesting/valuable. But I guess the financial risk of taking on OpenAI or Anthropic is quite high.
b) If you don't want bots scraping your content and DDOSing you, there are self-hosted alternatives to Cloudflare. The simplest one that I found is https://github.com/splitbrain/botcheck - visitors just need to press a button and get a cookie that lets them through to the website. No proof-of-work or smart heuristics.
This seems to be a totally normal sample size for such kinds of studies where you look at quantitative and qualitative aspects. Is this the only reason why you find the study to be bad?
I am a rather slow writer who certainly might benefit from something like Prism.
A good tool would encourage me, help me while I am writing, and maybe set up barriers that keep me from taking shortcuts (e.g. pushing me to re-read the relevant paragraphs of a paper that I cite).
Prism does none of these things - instead it pushes me towards sloppy practices, such as sprinkling citations between claims.
Why won't ChatGPT tell me how to build a bomb but Prism will happily fabricate fake experimental results for me?
According to the article, onboarding speed is measured as “time to the 10th Pull Request (PR).”
As we have seen on public GitHub projects, LLMs have made it really easy to submit a large number of low-effort pull requests without having any understanding of a project.
Obviously, such a kind of higher onboarding speed is not necessarily good for an organization.
reply