The attached post feels AI-generated/AI-edited. I don’t know why that detracts from the usefulness of this utility, but it — at least how I’m reacting to it — seems to.
For me, it reads as not having any soul to it, as well as changing who its target reader is throughout the post. The choice of emphasis comes across weirdly too.
I think this is a wonderful tool, but whether it’s because the author uses AI to help them write this post, or that there’s AI edited content everywhere now, I’m weary of reading and trusting something that reads like this post does.
I don’t know if this is overblown or meta-commentary on the state of online posts these days, but it’s how I’m feeling, more and more, when I see online writings.
The post is entirely authentic; it matches the writing style of the author from before the LLM boom and discusses work that the human author recently released. Can you pinpoint what makes you feel that way?
edit: I asked for explanation before the post was edited to expand on that. I disagree but am sympathetic to the weariness of reading content now that AI use is widespread.
What is this site? maarthandam.com? Is it a blog? An AI generated “newspaper”? An internet Newspaper? The menu doesn’t work on mobile, no articles appear to have a by-line, and there’s no link to outside sources to indicate the provenance of these quotes.
While there's not a lot of meat on the bone for this post, one section of it reflects the overall problem with the idea of Claude-as-everything:
> I spent weeks casually trying to replicate what took years to build. My inability to assess the complexity of the source material was matched by the inability of the models to understand what it was generating.
When the trough of disillusionment hits, I anticipate this will become collective wisdom, and we'll tailor LLMs to the subset of uses where they can be more helpful than hurtful. Until then, we'll try to use AI to replace in weeks what took us years to build.
If LLMs stopped improving today I’m sure you would be correct- as it is I think it’s very hard to predict what the future holds and where the advancements take us.
I don’t see a particularly good reason why LLMs wouldn’t be able to do most programming tasks, with the limitation being our ability to specify the problem sufficiently well.
I feel like we’ve been hearing this for 4 years now. The improvements to programming (IME) haven’t come from improved models, they’ve come from agents, tooling, and environment integrations.
> I feel like we’ve been hearing this for 4 years now.
I feel we were hearing very similar claims 40 years ago, about how the next version of "Fourth Generation Languages" were going to enable business people and managers to write their own software without needing pesky programmers to do it for them. They'll "just" need to learn how to specify the problem sufficiently well.
(Where "just" is used in it's "I don't understand the problem well enough to know how complicated or difficult what I'm about to say next is" sense. "Just stop buying cigarettes, smoker!", "Just eat less and exercise more, fat person!", "Just get a better paying job, poor person!", "Just cheer up, depressed person!")
Improved tooling/agent scaffolds, whatever, are symptoms of improved model capabilities, not the cause of better capabilities. You put a 2023-era model such as GPT-4 or even e.g. a 2024-era model such as Sonnet 3.5 in today's tooling and they would crash and burn.
The scaffolding and tooling for these models have been tried ever since GPT-3 came out in 2020 in different forms and prototypes. The only reason they're taking off in 2025 is that models are finally capable enough to use them.
Yet when you compare the same model in 2 different agents you can easily see capability differences. But cross (same tier) model in the same agent is much less stark.
My personal opinion is that there was a threshold earlier this year where the models got basically competent enough to be used for serious programming work. But all the major on the ground improvements since then has gone from the agents, and not all agents are equal, while all sota models are effectively.
> Yet when you compare the same model in 2 different agents you can easily see capability differences.
Yes definitely. But this is to be expected. Heck take the same person and put them in two different environments and they'll have very different performance!
> But cross (same tier) model in the same agent is much less stark.
Unclear what you mean by this. I do agree that the big three companies (OpenAI, Anthropic, Google DeepMind) are all more or less neck and neck in SOTA models, but every new generation has been a leap. They just keep leaping over each other.
If you compare e.g. Opus 4.1 and Opus 4.5 in the same agent harness, Opus 4.5 is way better. If you compare Gemini 3 Pro and Gemini 2.5 Pro in the same agent harness, Gemini 3 is way better. I don't do much coding or benchmarking with OpenAI's family of models, but anecdotally have heard the same thing going from GPT-5 to GPT-5.2.
The on the ground improvements have been coming primarily from model improvements, not harness improvements (the latter is unlocked by the former). Again, it's not that there were breakthroughs in agent frameworks that happened; all the ideas we're seeing now have all been tried before. Models simply weren't capable enough to actually use them. It's just that more and more (pre-tried!) frameworks are starting to make sense now. Indeed, there are certain frameworks and workflows that simply did not make sense with Q2-Q3 2025 models that now make sense with Q4 2025 models.
I actually have spent a lot of time doing comparisons between the 4.1 and 4.5 Claude models (and lately the 5.1->5.2 chatgpt models) and for many many tasks there is not significant improvement.
All things being equal I agree that the models are improving, but for many of the tasks I’m testing what has the most improvement is the agent. The agents choosing the appropriate model for the task for instance has been huge.
I do believe there is beneficial symbiosis but for my results the agent's provide much bigger variance than the model.
Both is true, models have also been significantly improved in the last year alone, let's not even talk about 4 years ago. Agents, tooling and other sugar on top is just that - enabling more efficient and creative usage, but let's not undermine how much better models today are compared to what was available in the past.
The code that's generated when given a long leash is still crap. But damned if I didn't use a JIRA mcp and a gitlab mcp, and just have the corporate AI just "do" a couple of well defined and well scoped tickets, including interacting with JIRA to get the ticket contents, update its progress, push to gitlab, and open an MR. Then, the corporate CodeRabbit does a first pass code review against the code so any glaring errors are stomped out before a human can review it. What's more scary though is that the JIRA tickets were created by a design doc that was half AI generated in the first place. The human proposed something, the AI asked clarifying questions, then broke the project down into milestones and then tickets, and then created the epic and issues on JIRA. One of my tradie friends taking an HVAC class tells me that there are a couple of programmers in his class looking to switch careers. I don't know what the future brings, but those programmers (sorry, "software developers") may have the right idea.
Yes we get it, there is a ton of "work" being done in corporate environments, in which the slop that generative AI churns out is similar to the slop that humans churn out. Congrats.
How do you judge model improvements vs tooling improvements?
If not working at one of the big players or running your own, it appears that even the APIs these days are wrapped in layers of tooling and abstracting raw model access more than ever.
> even the APIs these days are wrapped in layers of tooling and abstracting raw model access more than ever.
No, the APIs for these models haven't really changed all that much since 2023. The de facto standard for the field is still the chat completions API that was released in early 2023. It is almost entirely model improvements, not tooling improvements that are driving things forward. Tooling improvements are basically entirely dependent on model improvements (if you were to stick GPT-4, Sonnet 3.5, or any other pre-2025 model in today's tooling, things would suck horribly).
LLM capability improvement is hitting a plateau with recent advancements mostly relying on accessing context locally (RAG), or remotely (MCP), with a lot of extra tokens (read: drinking water and energy), being spent prompting models for "reasoning". Foundation-wise, observed improvements are incremental, not exponential.
> able to do most programming tasks, with the limitation being our ability to specify the problem sufficiently well
We've spent 80 years trying to figure that out. I'm not sure why anyone would think we're going to crack this one anytime in the next few years.
> Foundation-wise, observed improvements are incremental, not exponential.
Incremental gains are fine. I suspect capability of models scales roughly as the logarithm of their training effort.
> (read: drinking water and energy)
Water is not much of a concern in most of the world. And you can cool without using water, if you need to. (And it doesn't have to be drinking water anyway.)
Yes, energy is a limiting factor. But the big sink is in training. And we are still getting more energy efficient. At least to reach any given capability level; of course in total we will be spending more and more energy to reach ever higher levels.
Incremental gains in output seem to - so far - require exponential gains in input. This is not fine.
Water is a concern in huge parts of the World, as is energy consumption.
And if the big sink is “just” in training, why is there so much money being thrown at inference capacity?
I thought it was mad when I read that Bitcoin uses more energy than the country of Austria, but knowing AI inference using more energy than all the homes in the USA is so, so, so much worse given the quality of the outputs are so mediocre.
I would think/hope that the code assist LLMs would be optimizing towards supportable/legible code solutions overall. Mostly in that they can at least provide a jumping off point, largely accepting that they more often than not won't be able to produce complete, finished solutions entirely.
The folks that dismiss JB’s work by saying “this could have been done in <x>” are missing the point of why anything is done.
If you are entirely utilitarian in how you approach making a game (as in this case) then you’ll want to create as little as possible to make the game. An existing game engine, an existing programming language, existing libraries, etc.
If your goal is the economic return that making a game will (hopefully!) provide, this is understandable.
However, how I see JB based on his past work and talks is someone who wants to spend their life bringing things into existence. From all available evidence it appears the art of creating and the art of having created is his work and his legacy. The economic return is rhe by-product, but not the goal.
We are in this earth for a finite amount of years, and he is spending his time creating new things. It’s an admirable use of time, and at least from my perspective holds a universe of meaning that working under the utilitarian approach loses.
Blow made his own language because he's so eye-wateringly arrogant and thinks every language (that he didn't make) sucks, and only he is smart enough to design a better language for programming games.
Seriously, this is why he did it. His ego and arrogance is off the charts and if it wasn't made by him, he thinks it sucks (e.g. he doesn't like Linux, probably because he realizes Torvalds is actually smarter than him). He also doesn't like C++ or Rust, again, it's probably a good indicator he has a deep inferiority complex and so he has to prove he's the smartest person in the world by writing his own, "better" language.
I.e. I don't think he's making a programming language for some "love of creating", I think he's doing it because he has a deep psychological issue/insecurity, which drives his need to always be the "smartest person in the room", his arrogance, the way he dismisses others who don't agree with his viewpoints etc.
Even if you don't like Jon, calling Jai an exercise in arrogance is simply untrue. When he started making Jai in ~2014, there were very few viable alternatives to C/C++ in the systems programming space that offered the kind of expressive power becoming of a langauge built this century. Rust is great, but it prioritising correctness is not always the right choice, especially not for games. Jai introduced many ideas that languages like Zig and Odin ended up adopting.
It may not have a public* release but, over the last decade (starting pre-Zig/Odin), Blow has discussed it extensively in his videos[0], enough that even ~10y was possible for someone to make a toy independent implementation[1].
Still then, it's a stretch to say that Jai influenced other languages. How could it when only a handful of game-centered applications have been built by a handfull of devs?
Rust and Zig developed features by cutting their teeth on large amounts of real software, not by following one guy's personal project that has no source, no library, no spec available.
Jai, odin and zig's creators are all part of the handmade network, a community of programmers. You are vastly underestimating blow's reach/influence.
Odin's creator has credited Jai as an influence. You can see him in the comments of old jai youtube videos (videos that go into a lot of depth about the language design). Odin's syntax and features are very similar to Jai, the influence is pretty clear. Odin has other influences of course but you could say it's "jai but open source".
Lastly, jai is not open source but it doesn't mean it's not available. You can message blow to get access to it. Many programmers have used it. There are third party jai libraries on github.
I've never heard of Odin or seen any projects written in it, seen a company hire for it, or seen it discussed at a PL conference. There's no stable compiler for it, and no spec. Yeah, I'm just one person, so maybe I'm just in my own bubble, but these are hobby projects with a very small communities.
> Still then, it's a stretch to say that Jai influenced other languages. How could it when only a handful of game-centered applications have been built by a handfull of devs?
Lots of people have seen his talks about the language, so why do you think its impossible it influenced other languages?
It's unlikely that the Rust and Zig devs are looking at one guy's gamedev focused vlog compared to feedback from tens of thousands of engineers writing tens of thousands of public projects in Rust and Zig.
Have they heard of Jai? Yeah probably. But it's barely a drop in the bucket as far as the PL design community goes.
Oh, yes, the Rust team does "market research" and interviews people to see how they use the language, where the pain points are, etc. They have talks at Rustconf about how they gather information on how the language is used. Never seen them mention Jai.
> How has Jai introduce ideas if it’s not even released?
These are orthogonal concepts. Jai can or cannot introduce ideas, and Jai can or cannot be released. As of now, it is in fact so that Jai has introduced ideas, and has been released to a closed group of beta testers.
> How can we claim to know what it did “right” when only a few projects have been built in it?
To judge whether Jai did something right, in my opinion, it suffices to read the documentation and experience someone else programming second-hand and take advantage of its offerings, namely making programming less tedious, more enjoyable, more safe. It appears to me that you set the bar of usefulness or success too high for no good reason.
I've watched enough hours of his streams to know that this is NOT a reductive take. Blow is one of the most arrogant developers and game designers, and believes that nearly everyone else is an idiot.
He's somewhat Musk adjacent in his need to be viewed as smart (but I guess he does so least have way more programming chops than Musk, so I'll give him that).
C++ & Linux are world-changing tools, but C++ & Linux really do suck in ways that become more offensive with taste. Rust makes very different tradeoffs than ones gamedevs want.
Regardless, if arrogance drives people to make new tools then we should be grateful for that arrogance.
Think it's more along the lines of Jon having the ability to create a language, and upon being dissatisfied with what he was using, decided to make his own.
GitHub is littered with pet languages that people have made, and doubt their reasons are simply about being "eye-wateringly arrogant".
Moving past that, people paying attention or wanting to use the language, usually means it appeals to them. Jai has fans and supporters, because they are able to look past or are not concerned about his personality quirks, but are focused on the quality and usefulness of the software produced.
I think you're projecting a lot of your own complexes and insecurities.
He built a language for a very specific task: building games. There were quite a few requirements for such a language. Opinionated? Yes. But that's how you get new languages: by having opinions. Along the way he changed the design and the assumptions several times (e.g. built-in SOA structures are gone) while keeping the original goal in mind and using it to build a custom engine and a game while building the language (thus validating the choices made).
If/when Jai is released hopefully sometime next year, I do hope the documentation includes the rationale because he talked a lot about why other languages don't cut it in his opinion in the early days of development.
Eh, I can write that comment because it's fairly easy to see this side of JBlow if you've been following his work for a while. He is so naturally abrasive about other people's work, loves shitting on things he didn't make himself, loves being the smartest guy in the room, and also is a covid is a hoax, anti-vaxxer, Trump supporter etc.
I don't think I'm the smartest guy in the room, and that's OK. I realised a long time ago that ego/arrogance isn't a great quality and it's far better to have a strong network of friends and supporters, and that doesn't happen if you're an arrogant prick.
And yes, he built the language (which is totally un-needed) because the "idiots" who made all the existing languages, didn't make one as good as in JBlows brain. Despite the fact that there are 1000s of games which are far better than anything JB has made written in C#, C++, Java, Rust, etc. Did Larian need to write a new programming language to make Baldurs Gate 3?
Only JB is arrogant to think that only a new language is good enough for him to make a game with. A game that is just a modern spin on Sokoban and where he paid a bunch of other game devs to use their puzzles! You could write this shit in three.js and it wouldnt look or feel any differently.
+1 to all of this. I can no longer deal seriously with Blow's ideas, programming language, or games because he can't present any idea without being highly condescending and critical of just about everyone else. I'm glad I've never had to work for or with him, because he's the type of coworker or boss that constantly makes everyone's lives miserable.
Yes, you can do good things with shitty tools. And you could stop and say: this is enough. But then we would probably never have any programming languages at all.
Haskell exists because idiots that made existing languages didn't make one as good as in Philip Wadler's brain.
Go literally exists because idiots cannot use programming languages created by geniuses.
Rust exists because idiots who made all the existing languages didn't make one as good as in Graydon Hoare's brain. After all, all browsers on the market were built in C/C++, who is he to think that he could create a better/different language? Shut up and get on with the program.
C# exists because idiots who created other languages didn't create a language Microsoft wanted to control, and also weren't as good as the one Anders Hejlsberg's brain. After all, Java was already there.
Except Java exists only because who created other languages didn't create a language as good as the one in James Gosling's (and Mike Sheridan's and Patrick Naughton's) brain. Again, C/C++ had already been there, they could've used that.
Is Blow abrasive and shits on a lot of things? Of course. If you can't see past that to what he's actually doing with the language he's implementing, it's your problem.
> Only JB is arrogant to think that only a new language is good enough for him to make a game with.
Lol. I think this is the textbook definition of projection. He literally never said nor implied this in any way, shape, or form.
If anything, creating a new language set him back several years.
My main criticism of Blow is that he's consistently highly condescending to other games, game developers, and programmers. Many of whom have been shipping so many amazing and creative things while he's spent a decade making a Sokoban game.
> If your goal is the economic return that making a game will (hopefully!) provide, this is understandable
I don't know why the conversation always devolves into this. How it goes is "John cares about quality, everyone else only cares about money"
Choosing to prioritize art, story, and gameplay over raw execution speed does not mean you only care about money. It means you care about having a good game. That doesn't mean you can't do both, but if you have a time restriction, it's a completely reasonable trade off to make. Especially if your users won't even notice.
I would rather devs make games for people playing them, not for web devs who have Electron baggage.
One thing to point out that is lost in these arguments of “they create value for the shareholders”.
Folks that own a vast amount of stock do not pay taxes on that stock. They own the shares, and they take out loans against those shares. At some point they rollover or pay off those loans by selling some shares, but the shares have increased in value significantly in that time, or they’ve been granted new shares.
When we say “<business> has created value for shareholders”, it’s said in a way that implies that somehow that wealth creation makes its way into the tax system by virtue of the fact the wealth was ‘created’. It does not.
First, taxes still get paid when the individual dies as estate tax. Second, increased shareholder value typically means more corporate profit which is also taxed. Third, dividends are taxed. So your claim that the shareholder value never makes its way into the tax system is plainly false.
This is all aside from the fact that increased shareholder value means a more abundance society regardless of the increase in taxes. We could quibble over the exact distribution of who gains from the enlarged pie but it's certainly not the case the 100% of it goes to capitalists so consumers and employees also benefit.
> taxes still get paid when the individual dies as estate tax
Almost no one in the US pays the estate tax. It only applies to estates over $14MM and most large estates get reorganized into trusts with estate tax avoidance as a primary motive.
Yes this entire conversation is about the ultra wealthy not paying their "fair share". A $14MM exemption is practically irrelevant here.
> most large estates get reorganized into trusts with estate tax avoidance
This isn't so simple. Transfers to a irrevocable trust count against your lifetime 14mm estate and gift tax exemption and a trust in excess of the 14M exemption is subject to gift tax.
Also, this discussion was about "Buy Borrow Die" strategy. Irrevocable trusts don't make much sense in this context because trusts aren't subject to stepped up basis.
My biggest complaints about search come from day-to-day uses:
I use search in my email pretty heavily, and I’m most interested in specific words in the email; and when those emails are from specific folks or a specific domain. But, the mobile version of Gmail produces different results than the mobile Outlook app than the desktop version of Gmail, and all of them are pretty terrible at search as it pertains to email.
I have a hard to getting them to pull up emails in search that I know exist, that I know have certain words, and I know have certain email addresses in the body.
I recognize a generalized searching mechanisms is going to get domain specific nuances wrong, but is it really so hard to make a search engine that works on email and email based attachments that no one cares enough to try?
Huh, maybe your use case is around the indexing of the contents of attachments? I basically never search for the contents of attachments, just the clip does of emails, and have found gmail search to be really good. I switched back to the web client from Mac’s native mail app for this reason because search has been so good for me in Gmail.
I haven’t looked, but I wonder if there is a good hackable email client that will let you substitute out the search index with a reasonable abstraction from all the complicated email protocol stuff. I feel like building an index for your use case is totally achievable if so.
While we will never be able to get folks to stop using AI to “help” them shape their replies, it’s super annoying to have folks think that by using AI that they’re doing others a favor. If I wanted to know what an AI thinks I’ll ask it. I’m here because I want to know what other people think.
At this point, I make value judgments when folks use AI for their writing, and will continue to do so.
I strongly agree with this sentiment and I feel the same way.
The one exception for me though is when non-native English speakers want to participate in an English language discussion. LLMs produce by far the most natural sounding translations nowadays, but they imbue that "AI style" onto their output. I'm not sure what the solution here is because it's great for non-native speakers to be able to participate, but I find myself discarding any POV that was obviously expressed with AI.
If I want to participate in a conversation in a language I don't understand I use machine translation. I include a disclaimer that I've used machine translation & hope that gets translated. I also include the input to the machine translator, so that if someone who understands both languages happens to read it they might notice any problems.
When I occasionally use MTL into a language I'm not fluent in, I say so. This makes the reader aware that there may be errors unknown to me that make the writing diverge from my intent.
OTOH I am participating in a wonderful discord server community, primarily Italians and Brazilians, with other nationalities sprinkled in.
We heavily use connected translating apps and it feels really great. It would be such a massive pita to copy every message somewhere outside, having to translate it and then back.
Now, discussions usually follow the sun, and when someone not speaking, say, Portuguese wants to join in, they usually use English (sometimes German or Dutch), and just join.
We know it's not perfect but it works. Without the embedded translation? It absolutely wouldn't.
I also used pretty heavily a telegram channel with similar setup, but it was even better, with transparent auto translation.
Reddit would be even worse if the translations were better, now you don't have to waste much time because it hits you right in the face. Never ever translate something without asking about it first.
When I search for something in my native tongue it is almost always because I want the perspective of people living in my country having experience with X. Now the results are riddled with reddit posts that are from all over the world with crappy translation instead.
I think we should distinguish between the feature being good/hated:
1. An automatic translation feature.
2. Being able to submit an "original language" version of a post in case the translation is bad/unavailable, or someone can read the original for more nuance.
The only problem I see with #2 involves malicious usage, where the author is out to deliberately sow confusion/outrage or trying to evade moderation by presenting fundamentally different messages.
I think the audience that would be interested in this is vanishingly small, there exist relatively few conversations online that would be meaningfully improved by this.
I also suspect that automatically translating a forum would tend to attract a far worse ratio of high-effort to low-effort contributions than simply accepting posts in a specific language. For example, I'd expect programmers who don't speak any english to have on average a far lower skill level than those who know at least basic english.
That's Twitter currently, in a way. I've seen and had short conversations in which each person speaks their own language and trusts the other to use the built-in translation feature.
Indeed, this sort of “writing with an accent” can illuminate interesting aspects of both English and the speakers’ native language that I find fascinating.
100%! I will always give the benefit of the doubt when I see odd syntax/grammar (and do my best to provide helpful correction if it's off-base to the extent that it muddies your point), but hit me with a wordy, em-dash battered pile of gobbledygook and you might as well be spitting in my face.
Yep, it’s a 2 way learning street - you can learn new things from non native speakers, and they can learn from you as well. Any kind of auto Translation removed this. (It’s still important to have for non fluent people though!)
I honestly think that very few people here are completely non-conversant in English. For better or worse, it's the dominant language. Amost everyone who doesn't speak English natively learns it in school.
I'm fine with reading slightly incorrect English from a non-native speaker. I'd rather see that than an LLM interpretation.
...I'm not sure I agree. I sometimes have a lot of trouble understanding what non-English speakers are trying to say. I appreciate that they're doing their best, and as someone who can only speak English, I have the utmost respect anyone who knows multiple languages—but I just find it really hard.
Some AI translation is so good now that I do think it might be a better option. If they try to write in English and mess up, the information is just lost, there's nothing I can do to recover the real meaning.
The solution is to use a translator rather than a hallucinatory text generator. Google Translate is exceptionally good at maintaining naturalness when you put a multi-sentence/multi-paragraph block through it -- if you're fluent in another language, try it out!
Google translate used to be the best, but it's essentially outdated technology now, surpassed by even small open-weight multilingual LLMs.
Caveat: The remaining thing to watch out for is that some LLMs are not -by default- prompted to translate accurately due to (indeed) hallucination and summarization tendencies.
* Check a given LLM with language-pairs you are familiar with before you commit to using one in situations you are less familiar with.
* always proof-read if you are at all able to!
Ultimately you should be responsible for your own posts.
I haven't had a reason to use Google Translate in years, so will ask: Have they opted to not use/roll out modern LLM translation capabilities in the Google Translate product?
I have seen Google Translate hallucinate exactly zero times over thousands of queries over the years. Meanwhile, LLMs emit garbage roughly 1/3 of the time, in my experience. Can you provide an example of Translate hallucinating something?
Agreed, and I use G translate daily to handle living in a country where 95% of the population doesn’t speak any language I do.
It occasionally messes up, but not by hallucinating, usually grammar salad because what I put into it was somewhat ambiguous. It’s also terrible with genders in Romance languages, but then that is a nightmare for humans too.
Hard disagree. Google Translate performance is abysmal when dealing with danish. In many cases its output is unusable. On the other hand, ChatGPT is excellent at it.
IMO chatgpt is a much better translator. Especially if you’re using one of their normal models like 5.1. I’ve used it many times with an obscure and difficult slavic language that i’m fluent in for example, and chatgpt nailed it whereas google translate sounded less natural.
The big difference? I could easily prompt the LLM with “i’d like to translate the following into language X. For context this is a reply to their email on topic Y, and Z is a female.”
Doing even a tiny bit of prompting will easily get you better results than google translate. Some languages have words with multiple meanings and the context of the sentence/topic is crucial. So is gender in many languages! You can’t provide any hints like that to google translate, especially if you are starting with an un-gendered language like English.
I do still use google translate though. When my phone is offline, or translating very long text. LLM’s perform poorly with larger context windows.
Maybe they should say "AI used for translation only". And maybe us English speakers who don't care what AI "thinks" should still be tolerant of it for translations.
I have found that prompting "translate my text to English, do not change anything else" works fine.
However, now I prefer to write directly in English and consider whatever grammar/ortographic error I have as part of my writing style. I hate having to rewrite the LLM output to add myself again into the text.
one solution that appeals to me (and which i have myself used in online spaces where i don't speak the language) is to write in a language you can speak and let people translate it themselves however they wish
i don't think it is likely to catch on, though, outside of culturally multilingual environments
TL;DR: Ask for a line edit, "Line edit this Slack message / HN comment." It goes beyond fixing grammar (because it improves flow) without killing your meaning or adding AI-isms.
When I hear "ChatGPT says..." on some topic at work, I interpret that as "Let me google that for you, only I neither care nor respect you enough to bother confirming that that answer is correct."
To my mind, it's like someone saying "I asked Fred down at the pub and he said...". It's someone stupidly repeating something that's likely stupid anyway.
You can have the same problem with Googling things, LLMs usually form conclusions I align with when I do the independent research. Google isn't anywhere near as good as it was 5 years ago. All the years of crippling their search ranking system and suppressing results has caught up to them to the point most LLMs are Google replacements.
In a work context, for me at least, this class of reply can actually be pretty useful. It indicates somebody already minimally investigated a thing and may have at least some information about it, but they're hedging on certainty by letting me know "the robots say."
It's a huge asterisk to avoid stating something as a fact, but indicates something that could/should be explored further.
(This would be nonsense if they sent me an email or wrote an issue up this way or something, but in an ad-hoc conversation it makes sense to me)
I think this is different than on HN or other message boards, it's not really used by people to hedge here, if they don't actually personally believe something to be the case (or have a question to ask) why are they posting anyway? No value there.
> can actually be pretty useful. It indicates somebody already minimally investigated a thing
Every time this happens to me at work one of two things happens:
1) I know a bit about the topic, and they're proudly regurgitating an LLM about an aspect of the topic we didn't discuss last time. They think they're telling me something I don't know, while in reality they're exposing how haphazard their LLM use was.
2) I don't know about the topic, so I have to judge the usefulness of what they say based on all the times that person did scenario Number 1.
Yeah if the person doing it is smart I would trust they had the reasonable prompt and ruled out flagrant BS answers. Sometimes the key thing is just to know the name of the thing for the answer. It's equally as good/annoying as reporting what Google search gives for the answer. I guess I assume mostly people will do the AI query/search and then decide to share the answer based on how good or useful it seems.
These days, most people who try googling for answers end up reading an article which was generated by AI anyway. At least if you go right to the bot, you know what you're getting.
> When I hear "ChatGPT says..." on some topic at work, I interpret that as "Let me google that for you, only I neither care nor respect you enough to bother confirming that that answer is correct."
I have a less cynical take. These are casual replies, and being forthright about AI usage should be encouraged in such circumstances. It's a cue for you to take it with a grain of salt. By discouraging this you are encouraging the opposite: for people to mask their AI usage and pretend they are experts or did extensive research on their own.
If you wish to dismiss replies that admit AI usage you are free to do so. But you lose that freedom when people start to hide the origins of their information out of peer pressure or shame.
If someone is asking a technical question along the lines of “how does this work” or “can I do this,” then I’d expect them to Google it first. Nowadays I’d also expect them to ask ChatGPT. So I’d appreciate their preamble explaining that they already did that, and giving me the chance to say “yep, ChatGPT is basically right, but there’s some nuance about X, Y, and Z…”
Expecting people to stop asking casual questions to LLMs is definitely a lost cause. This tech isn't going anywhere, no matter how much you dislike it.
> expecting anyone to actually try anymore is a lost cause
Well now you're putting words in my mouth.
If you make it against the rules to cite AI in your replies then you end up with people masking their AI usage, and you'll never again be able to encourage them to do the legwork themselves.
But I'm not interested in the AI's point of view. I have done that myself.
I want to hear your thoughts, based on your unique experience, not the AI's which is an average of the experience of the data it ingested. The things that are unique will not surface because they aren't seen enough times.
Your value is not in copy-pasting. It's in your experience.
Did you agree with it before the AI wrote it though (in which case, what was the point of involving the AI)?
If you agree with it after seeing it, but wouldn't have thought to write it yourself, what reason is there to believe you wouldn't have found some other, contradictory AI output just as agreeable? Since one of the big objections to AI output is that they uncritically agree with nonsense from the user, scycophancy-squared is even more objectionable. It's worth taking the effort to avoid falling into this trap.
Well - the point of involving the AI is that very often it explains my intuitions way better than I can. It instantiates them and fills in all the details, sometimes showing new ways.
I find the second paragraphs contradictory - either you fear that I would agree with random stuff that the AI writes or you believe that the sycophant AI is writing what I believe. I like to think that I can recognise good arguments, but if I am wrong here - then why would you prefer my writing from an LLM generated one?
> Well - the point of involving the AI is that very often it explains my intuitions way better than I can. It instantiates them and fills in all the details
> I like to think that I can recognise good arguments, but if I am wrong here - then why would you prefer my writing from an LLM generated one?
Because the AI will happily argue either side of a debate, in both cases the meaningful/useful/reliable information in the post is constrained by the limits of _your_ knowledge. The LLM-based one will merely be longer.
Can you think of a time when you asked AI to support your point, and upon reviewing its argument, decided it was unconvincing after all and changed your mind?
You could instead ask Kimi K2 to demolish your point instead, and you may have to hold it back from insulting your mom in the ps.
Generally if your point holds up under polishing under Kimi pressure, by all means post it on HN, I'd say.
Other LLMs do tend to be more gentle with you, but if you ask them to be critical or to steelman the opposing view, they can be powerful tools for actually understanding where someone else is coming from.
Try this: Ask an LLM to read the view of the person you're answering to, and ask it steelman their arguments. Now think to see if your point is still defensible, or what kinds of sources or data you'd need to bolster it.
> why would you prefer my writing from an LLM generated one?
Because I'm interested in hearing your voice, your thoughts, as you express them, for the same reason I like eating real fruit, grown on a tree, to sucking high-fructose fruit goo squeezed fresh from a tube.
"I asked an $LLM and it said" is very different than "in my opinion".
Your opinion may be supported by any sources you want as long as it's a genuine opinion (yours), presumably something you can defend as it's your opinion.
If I wanted to consult an AI, I'd consult an AI. "I consulted an AI and pasted in its answer" is worse than worthless. "I consulted an AI and carefully checked the result" might have value.
Some will blindly dismiss anything using them as AI generated, but realistically the em-dash is only one sign among many. Way more obvious is the actual style of the writing. I use Claude all of the time and I can instantly tell if a blog post I’m reading was written with Claude. It is so distinctive. People use some of the patterns it uses some of the time. But it uses all of them all of the time.
I think there's well done and usually unnoticeable and poorly done and insulting. I don't agree that the two are always the same, but I think lots of people might think they are doing the former but are not aware enough to realize they are doing the latter.
When someone says: "Source?", is that kinda the same thing?
Like, I'm just going to google the thing the person is asking for, same as they can.
Should asking for sources be banned too?
Personally, I think not. HN is better, I feel, when people can challenge the assertions of others and ask for the proof, even though that proof is easy enough to find for all parties.
IMO, HN commenters used to at least police themselves more and provide sources in their comments when making claims. It was what used to separate HN and Reddit for me when it came to response quality.
But yes it is rude to just respond "source?" unless they are making some wild batshit claims.
I actually use LLMs to help me dig up the sources. It's quicker than google and you get them nicely formatted besides.
But: Just because it's easy doesn't mean you're allowed to be lazy. You need to check all the sources, not just the ones that happen to agree with your view. Sometimes the ones that disagree are more interesting! And at least you can have a bit of drama yelling at your screen at how dumb they obviously are. Formulating why they are dumb, now there's the challenge - and the intellectual honesty.
But yeah, using LLMs to help with actually doing the research? Totally a thing.
I think what's important here is to reduce harm even if it's still a little annoying. Because if you try to completely ban mentioning something is LLM written you'll just have people doing it without a disclaimer...
Yes, comments of this nature are bad, annoying, and should be downvoted as they have minimal original thought, take minimal effort, and are often directly inaccurate. I'd still rather they have a disclaimer to make it easier to identify them!
Further, entire articles submitted to HN are clearly written by a LLM yet get over a hundred upvotes before people notice whether there's a disclaimer or not. These do not get caught quickly, and someone clicking on the link will likely generate ad revenue that incentives people to continue doing it.
LLM comments without a disclaimer should be avoided, and submitted articles written by a LLM should be flagged ASAP to avoid abuse since by the time someone clicks the link it's too late.
What LLM generate is an amalgamation of human content they have been trained on. I get that you want what actual humans think, but that’s also basically a weighted amalgamation. Real, actual insight, is incredibly rare and I doubt you see much of it on HN (sorry guys; I’ll live with the downvotes).
Why do you suppose we come to HN if not for actual insight? There are other sites much better for getting an endless stream of weighted amalgamations of human content.
Agree and I think it might also be useful to have that be grounds for a shadowban if we start seeing this getting out of control. I'm not interested, even slightly, in what an LLM has to say about a thread on HN. If I see an account posting an obvious LLM copy/paste, I'm not interested in seeing anything from that account either. Maybe a warning on the first offense is fair, but it should not be tolerated or this site will just drown in the slop.
when it had only two comments. One of them was the Gemini summary, which had already been massively downvoted. I couldn't make heads or tails of the paper posted, and probably neither could 99% of other HNers. I was extremely happy to see a short AI summary. I was on my phone and it's not easy to paste a PDF into an LLM.
When something highly technical is posted to HN that most people don't have the background to interpret, a summary can be extremely valuable, and almost nobody is posting human-written summaries together with their links.
If I ask someone a question in the comments, yes it seems rude for someone to paste back an LLM answer. But for something dense and technical, an LLM summary of the post can be extremely helpful. Often just as helpful as the https://archive.today... links that are frequently the top comment.
When there's nothing else to go on, it's still more useful than nothing.
The story was being upvoted and on the front page, but with no substantive comments, clearly because nobody understood what the significance of the paper was supposed to be.
I mean, HN comments are wrong all the time too. But if an LLM summary can at least start the conversation, I'm not really worried if its summary isn't 100% faithful.
That's a pretty good example. The summary is actually useful, yet it still annoys me.
But I'm not usually reading the comments to learn, it's just entertainment (=distraction). And similar to images or videos, I find human-created content more entertaining.
One thing to make such posts more palatable could be if the poster added some contribution of their own. In particular, they could state whether the AI summary is accurate according to their understanding.
I definitely read the comments to learn. I love when there's a post about something I didn't know about, and I love when HN'ers can explain details that the post left confusing.
If I'm looking for entertainment, HN is not exactly my first stop... :P
Most often I see these answers under posts like "what's the longest river or earth", or "is Bogota a capital of Venezuela?"
Like. Seriously. It often takes MORE time to post this sort of lazy question than actually look it up. Literally paste their question into $search_engine and get 10 the same answers on the first page.
Actually sometimes telling a person like this "just Google it" is beneficial in two ways: it helps the poster develop/train their own search skills, and it may gently nudge someone else into trying that approach first, too. At the same time slowing the raise of the extremely low effort/quality posts.
But sure, sometimes you get the other kind. Very rarely.
I’ve seen so many SO and other forum posts where the first comment is someone smugly saying “just google it, silly”.
Only that, I’m not the one who posted the original question, I DID google (well DDG) it, and the results led me to someone asking the same question as me, but it only had that one useless reply
Or worse, you google an obscure topic and the top reply is “apple mountain sleep blue chipmunk fart This comment was mass deleted with Redact” and the replies to that are all “thanks that solved my problem”
Agreed, with a caveat. If someone is asking for an objective answer which could be easily found with a search, and hasn't indicated why they haven't taken that approach, it really comes across as laziness and offloading their work onto other people. Like, "what are the best restaurants in an area" is a good question for human input; "how do you deserialize a JSON payload" should include some explanation for what they've tried, including searches.
But if I wanted to ask an AI I would put that into ChatGPT, not ask HN. I would only ask that on HN if I wanted other people's opinions!
You could reply with "Hey you could ask [particular LLM] because it had some good points when I asked it" but I don't care to see LLM output regurgitated on HN ever.
I strongly disagree - when I post something that AI wrote I am doing it because it explains my thoughts better than I can - it digs deeper and finds the support for intuitions that I cannot explain nicely. I quote the AI - because I feel this is fair - if you ban this you would just lose the information that it was generated.
I thought that the point was to post valuable thoughts - because it is interesting to read them. But now you suggest that it depends on how they were generated.
Yeah, but if you're having to turn to a machine to compose your thoughts on a subject they're probably not that valuable. In an online community like this the interesting (not necessarily valuable) thoughts are the ones that come from personal experience, and raise the non-obvious points that an LLM is never going to come up with.
The "They" here are the folks who are currently investing in 'selling' AI solutions to other companies. OpenAI, Microsoft, Google's Gemini, and a slew of AI-backed startups are good examples.
They don't need AI to turn a profit.
They need AI to be seen as widely adopted and "a part of life".
They need certain categories of folks (CEOs, CIOs, Boards of Directors) to see AI as valuable enough to invest in.
They need to keep up the veneer of success long enough to make their investments attractive to acquisition by Private Equity or to an IPO.
They need to juice the short-term stock price.
Their goal isn't to produce a long-term business, their goal is to raise short-term returns to the point that the investors get a nice return on their investment, and then it becomes someone else's problem.
"How money works" YouTube channel had a nice video about this trend in particular going back to making stock buybacks legal in 1982 I think, which made CEO and execs wealth acquisition driven not by a long successful career with healthy margins and dividends, but a short-tenured local maximum pump-and-dump and a hold-the-bag game funded by endless fiat currency which is printed on the backs of other people. Other people's money , Gordon Gecko, they're not just real, they're celebrity sociopaths running us into the ground because of a fragile ego.
On the other hand some notable open source leaders seem to be abrasive assholes. Linus, Theo, DHH, just three examples who come to mind. I think if you have a clear vision of what you want your project to be then being agressively dismissive of ideas that don't further that vision is necessary just to keep the noise to a low roar.
For me, it reads as not having any soul to it, as well as changing who its target reader is throughout the post. The choice of emphasis comes across weirdly too.
I think this is a wonderful tool, but whether it’s because the author uses AI to help them write this post, or that there’s AI edited content everywhere now, I’m weary of reading and trusting something that reads like this post does.
I don’t know if this is overblown or meta-commentary on the state of online posts these days, but it’s how I’m feeling, more and more, when I see online writings.
reply