Yeah Gemini has been consistently better than open ai's chat in icelandic, but I would still say it's far from passable as natural sounding. Lots of grammar errors and the pronunciation sounds like a non native speaker.
This depends on what you are building and what your goals are. What pg is referring to is how to build transformative companies, unicorns if you will.
There's plenty of businesses where you are simply optimizing a funnel, and driving more revenue at all costs is fully aligned with the goals of the company
I'm nearly always building early stage companies, where even the dev and operational overhead of charging for some added extra is not guaranteed to be paid for by the money collected.
The real reason is that it is not in the ai companies' best interest for the ais to be fair and truthful. They stand to gain from having the most dangerous or most deceiving ai, and this the most valuable
agreed to some extent. I think this parody still highlights what I feel is often the experience. It might not happen on a simple task such as changing a button color, but on more complicated things, this can definitely be exactly what it feels like.
I believe this is referencing a meme on the primeagen's standup podcast, where Casey is referred to as legendary while he feels undeserving of this title.
Titles aside, his talk is really insightful and it is super interesting to do a deep dive on these old computer/programming topics as the modern concepts were being discovered
Hah, I wasn't aware of that. I was sort of referring to what seems to be not such a meme that jblow is always introduced like that. It just seems weird, even if true...
Yeah Ive been meaning to watch that talk - I love listening to pretty much anything Casey says/does. He's extremely thoughtful and fair.
Thank you for bringing a term to a thing that has been bugging me: people who purport to be devs but spend most of their time on podcasts and blogs talking and writing about being devs, rather than just being devs.
This article is highly AI generated. The first paragraph is probably not AI generated, but the rest is definitely.
I also double checked on gptzero me and it 100% agrees with me.
I'm curious to know though whether or not the "author" used Gemini. I've been using Gemini a lot in the past year, and the writing sounds exactly like Gemini. But it's possible that all the models sound the same.
Edit: I'd like to add that I still liked the article and agree with it, and in general I find Gemini's writing style to be quite enjoyable.
How do you know? I think LLMs are more than advanced enough to come up with this text.
In fact, I think they've got so much more management in their training set than a senior engineer that I expect they're going to be a better source of material than the author.
The title was already a clue that it's AI-generated, and it only gets worse...
The first paragraph is probably not AI generated
"A bias toward action is a superpower only when applied correctly - here's how to frame it as moving decisively after building context, not rushing in before you have it."
...followed by many, many bloated words that can basically be summed up in two words of old Internet lore: LURK MOAR
I agree with the article too, but I didn't have to read that many words to know. "Don't just start doing things because you want to show off, and understand what value you're providing" is how I'd phrase it.
The images are awesome, in an “I cannot believe someone would upload this to their site” kind of way.
The blueprint with the “LOAD-BEARING WALL!!!” includes nonsensical labels like “Dining Broom” and “Witchen”, and I am supremely disappointed that there isn’t a bubbling cauldron or pointed hat to be found.
I don’t see why it is interesting to claim that a piece is AI generated, especially in light of the concession in your last paragraph that you liked the article.
AI is a tool. It outputs things only when asked to. Asking it requires skill. Saying “This is AI generated” is like pointing at furniture and saying that “power tools were used to build that” or at a program and saying “that’s not assembly, a modern garbage-collected language was used to build that” or at a meal and saying “this was cooked in a kitchen with a modern temperature-controlled oven” or at a paper and saying “this was obviously typeset in LaTeX, not roff or plain TeX.”
So what? Give me a shop full of power tools and I can’t build any furniture, give me Python and a bunch of libraries and I can’t write Mercurial, and give me any LLM and I couldn’t have written this piece.
Does the conceit come from the idea of “some LLM wrote this, so anyone could”? This is like walking through a modern art museum and seeing some obvious-looking piece and saying “I could have done that.” Well, you didn’t.
> AI is a tool. It outputs things only when asked to.
This is not entirely accurate if we consider agentic loops.
> Asking it requires skill.
Considering the realm of writing a blog post instead of programming, what is the skill? I suspect there is extremely little. But I could be wrong, tell me what skill you think went into this.
> and give me any LLM and I couldn’t have written this piece.
Prompt your LLM of choice to generate 100 articles about different aspects of joining a new role, then you won’t need to provide the insights into the specific situations the author described. All you need to do is pay for the tokens.
> Does the conceit come from the idea of “some LLM wrote this, so anyone could”?
Yes.
> Well, you didn’t.
Do you know that? What’s stopping someone from asking an LLM to generate an entire HN front page worth of articles, and reading that list instead of HN?
This really is a problem with the health of the site in my opinion. If there are a flood of AI generated articles in the top spot, the value proposition of HN as a site is diminished.
Hey, thanks for taking the time to respond to feedback.
I believe there is a societal problem here, not specific to your particular actions. The expectation that communications should be “cleaned up” and “professional” in the first place comes from a bygone era. In the 2010s and earlier removing all typos, fixing wording, formatting the text and adding images were all signals that someone put extra care into sending their message. But in the present that just indicates use of AI.
You mentioned in another comment that your hand-drawn images would be “far worse”. Leaving aside whether any images were needed at all, a hand-drawn scribble in the 2010s might be unprofessional but in the present day there’s an AI generated image on every visible surface. A hand-drawn image, even a bad one, would actually make you stand out.
Even if you want to present the cleaned up version, you could also provide the original audio.
Appreciate the feedback and I don't necessarily disagree with your points - the societal argument is an interesting one.
I do think it's interesting that some people are immediately turned off of something because of the whiff of AI even if they like the actual idea or content.
Will likely experiment with different ways of writing in the future!
broadly speaking AI has shown itself to be unreliable in a lot of domains so even if its strongest feature is in fact language, it still carries a readers predisposed experiences with the tech.
That is just one problem. You also have to contend with the onslaught of AI generated content that is being used in deliberately fake articles, online personas and the mountains and mountains of things people and entities are trying to push to make a buck. Scams, misinformation, state-sponsored propaganda, a lot of things that are arguably bad for society. So if you consider that, you must also accept that you are making a choice to use a linguistic style that is used by many websites that are vectors that do harm or deliberately mislead (which also does harm).
Even if the ideas are of your own merit, you can't expect readers to always do the literary surgery it will take to identify that.That's extra investigative work and assessment that was never a part of anyones reportoire before all this took off. But at this point i don't think you should be surprised that AI writing isn't met with open arms, AI writing has become a pain point for a lot of people in their personal lives and their work lives, a lot of organizations are wounding themselves as people throw their ideas around curated in AI and everyone's eyes are glazing over. It's become synonymous with "this guy does not understand the claude proposal he just sent me". And the choice is yours if you still want to use it knowing that, I just don't think you should be surprised if it doesn't get the reception you hoped for.
Plus, you should lean into whatever your monologue is to write. Even if you're afraid it's not pro enough, like. That's writing. It's a journey it has crossover with your ability to communicate it defines you as a person. Keep sharpening that tool and you have untold ways to convey ideas exactly the way you want people to experience them for the first time.
Why would I not just type in "how do I correctly balance towards action in a new role" into chatgpt and read that instead? I have no signal that the author did anything beyond that himself.
How would I discern that from the AI slop images and AI formatting? What you just said is not signal, it's things you privately know. There are a lot of blogs out there. You need to appear superficially worth reading before I know your biography and work processes.
"You need to..." implies "for me to read it", me possibly being a proxy for others, and nothing more. So responding to "You need to do X for me to read it" with " I don't need to do anything. You can choose to read or choose not to read." is just restating my comment tautologically. I think you are taking my criticism personally, but it's not personal or even related to the content because I don't know you and I didn't read it. I am just giving you the feedback that AI writing is a negative signal that is going to make a lot of people not read the thing, not because it's annoying to read, but just because there's no inherent indication that anyone thought through this text like there usually is when something is published.
Also, you engaged with my comment, not I with yours, and I attempted to converse with you, but you then exited superciliously (below) without engaging with or even acknowledging my point. You don't need to say goodbye online if you don't want to discuss what I'm saying.
I remember long ago, when I first met a computer person in college and he told me about PCs vs UNIX, I asked him, "why don't they make it run on a PC?"
it's amazing from a historical perspective that there are HNers around now who don't know who DHH is, when he was such a prevailing voice in the HN scene back in the day.
He is the author of rails and there are so many founders who were inspired by 37 signals and their takes on things back in the day.
I have trouble imagining it. The early rails community was very welcoming at the time. The things he now openly states were simply not welcome in most polite society at the time.
I'm not sure DHH really held the same views back then, either. The "Overton Window" has moved over the last 20 years so openly bigoted and racist views are accepted in the discourse now.
Marc Andreessen helped establish the modern web browser, he has always been an Ayn Rand fan for example. But politics wasn't that important back then, I guess? And I just think there simply were not a lot of opportunities to gain some political and financial capital from voicing their fringe believes for people like them back in the day. They were busy building systems and software, now they are busy building or changing political systems.
Libertarian views have always been relatively commonplace in the tech scene. Being a fan of Ayn Rand probably put you in the majority, but left-libertarian views were commonplace, too.
Calling for ethnic cleansing is not a libertarian belief. It is a fascist belief. "Privately owned roads are a good idea" and "Let's round up all the gypsies" are not equivalently debatable ideas.
Personality have needed for the following reasons (when wanting to preserve history): removing personal info, removing sensitive keys/passwords that were accidently committed a long time ago, splitting a repo. There's lots of use cases
I support anti-dox/and security practices by cleaning up repo history and rotating where you can, as long as your safe with the assumption that anyone with a copy of the history could push it back to central at any point.
But I really don't like the trend of some folks "need a clean history" and then disabled merge commits.
It's like they are saying "cool snap shot system you got there, id hate for it to be use to take a bunch of incremental snap shots with specific notes attached to each one. Why don't you let me bundle those all together real purdy like and no one has to know it took you 20 commits to fix that decade old bug."
I have literally never heard a reasonable claim that needing a "clean main branch" is more useful in a retro or bisect.
>"cool snap shot system you got there, id hate for it to be use to take a bunch of incremental snap shots with specific notes attached to each one. Why don't you let me bundle those all together real purdy like and no one has to know it took you 20 commits to fix that decade old bug."
It’s more like saying “in any writing process, revising is going to lead to better finished output.” It is absolutely true that no one needs to know that it took you 20 commits to fix that bug. It’s much easier to review 3 commits, one of which contains the refactoring done as part of the fix with the other two each containing changes that address distinct pieces of the bug. The structure of the change is clearer, and the structure of the attached notes will be clearer as well.
I don't think all writing process produces the same kinda of output, the editing of code itself is metadata I find more valuable and an artificially straight line.
Same thing I was thinking. I personally don't track unique visitors on my blog, just random sessions where you can't identify unique users, which allows me to not show a banner. If I were to use this technique I'd have to show a banner
reply