It is, indeed, justa tool. Fortunately, no one here is arguing that LLMs have consciousness.
I judge the output, but in the output I also include a lot of externalities: the electricity and water use for the datacenters. The increased website traffic by the scrapper bots. And also the deeper consequences on how we work. Working without LLMs forced us to find ways to collaborate: help newcomer learn things, have experienced people write documentation, talk tg each other to build a common understanding of things, etc. For me, this is a very important part of Haiku, and of the open internet in general.
I feel that LLMs are going to change that. Already the traffic by scrapper bots forced us to put some resources behind login walls. Search engines are going to return an LLM summary instead of sending you to an actual website for answers, which disconnects the answer from the place and heople who made it. You can do an entire project without talking to anyone but your LLM.
There is a possibility here that all this results in isolating people from each other, preventing collective organization, if itâs used in this way. I do make a lot of my work available under open source licenses. The only profit I get from that is the occasional âthank youâ email, people recognizing my nickname in unexpected places, things made using my software. I see a future where people will not access this information directly, but through an LLM mixing it with other data, and providing better answers, but no human connexions.
And all of this, under the control of a few US corporations who were already in control of way too many things.
For me, all of this is the output of the LLM tool. And even if it allows people to write software better, I donât think itâs worth losing all that.
I am one of those who support the moderatorsâ views on LLMs and the actions they have taken. I havenât felt it necessary to make a thousand posts explaining my position, but I am doing so now so that the LLM side donât assume that the silence of people like me means that we donât fully support the moderation team and their stance on AI.
I donât always agree with Pulkomandy (on other matters) but it would be tragic if he were to reduce his involvement in Haiku (and in the forum especially) because of the rudeness of some people on here who should know better. It would be hard to think of anybody who has contributed more to the project, and in particular to explaining it.
Sadly, I feel that the gulf between the AI brigade abd the non-AI brigade is so great that a fork of Haiku might be the best way forward. Perhaps the only way. We canât go on like this.
Iâve noticed an interesting pattern in this discussion, and frankly, in all similar debates about AI online.
When people speak out against AI, they almost always talk about large-scale global problems: climate change, large corporations, carbon footprints, or the loss of human connection.
But when people defend the use of AI, their arguments are always personal and practical. They share real-life stories: âI used it to translate my thoughts into another language so I could participate in discussions,â âI finally wrote a missing driver for a device,â or âI saved time on a boring task.â
Donât get me wrong - global problems like the environment and corporate monopolies may exist. But letâs be realistic: solving them is completely beyond the scope and purpose of the Haiku project. Weâre here to build an operating system, not save the world.
When we try to combat global problems with our forum rules, weâre simply blocking real people who are trying to do their best to patch holes where real developers lack the time, energy, or knowledge. Letâs focus on our real goal - writing good code and creating a great OS - instead of trying to save the world here.
If you do, you will surely be aware that your one vote wonât change anything, but if lots of people vote as you do, then you will change something.
Of course Haiku alone wonât change anything, but thatâs no argument for joining those who use AI and thereby contribute to the vast environmental damage it creates.
@Sebrof, regarding your analogy about elections - I live in a part of the world where people have a very, letâs say, pragmatic understanding of how much a single vote can change things. So the analogy doesnât quite work for me.
But returning to tech and ecology: using a pre-trained LLM for a few minutes to generate a missing driver uses roughly the same amount of electricity as boiling a kettle. I would argue that keeping our PCs running for hours just to argue about this on the forum leaves a much bigger carbon footprint.(A comparative study of AI and human programming on environmental sustainability - PMC)
If we truly want to save the environment, maybe we should focus on making Haiku more power-efficient on modern hardware, rather than policing how developers write their code?
This is the most interesting one to talk about of the ones you mentioned, because itâs not just some abstract âlarge-scale global problemâ. The reduction in human interaction from the use of LLMs can be measured on the small scale, too. How many people in the Haiku community choose to interact with their LLMs instead of asking questions and talking about their work, not just on these forums but on any communication medium?
I live in a part of the world where people didnât use to have that âpragmatic understandingâ you speak of, but it seems more and more people have it in recent years. And in many cases, I donât disagree.
But a vote is something you do at most once every few years, and it takes just seconds. Ideally those seconds would be the ones that matter, but if they donât, is it really so surprising? What surely has much more of an effect is whatever we choose to do, day in and day out, year in and year out. And that effect will be on ourselves, and the people around us, and the communities we are actually a part of.
And the world will change if the people in it change. âWhat would the world look like if everyone acted in this or that way?â is a question I think we should consider more often than we do. Perhaps we know for a fact that the vast majority of people wonât act in this or that way, and that may be tragic if they act in some other, worse way. But we canât affect those abstract, distant people; we can affect ourselves and the people we interact with.
LLMs and their usage is clearly erosive of many human goods, and most uses actively supplant human intelligence. Many people donât seem to care about those human goods or the use of human intelligence, but rather care more about some other metric, like how many new features their software has, or how easy it was to do some particular task, or whatever. But I am not here to build a world for machines; I am here to build a world for people. Haiku has always prioritized things other than âhow fast can we get things done?â and âhow many new features can we get in?â.
I think one of the key words is right there in the description of what Haiku is thatâs on our homepage:
Haiku is an open-source operating system that specifically targets personal computing.
You could define âpersonal computingâ, I suppose, to mean âany kind of computing a person would want to doâ. I think that is too broad a definition; if that really were the definition, well, âheadless server computingâ can be done by a person, too, right? Rather I think it ought to be defined more narrowly: the sort of computing a person does in harmony with the rest of their life and work. Not a tool to replace parts of human capacities, but to work with them. And if âAIâ wasnât about âreplacingâ, well â then why is the word âintelligenceâ right there in its name?
So if the primary argument for LLMs is âthey get more things done and fasterâ, and the primary argument against LLMs is âthey erode human goodsâ, well, is it any surprise that I am opposed?
We live in the world driven by the law of evolution. Everything is driven by it: Universe itself, planet system, biological species, humans and society, technology. More adapted ones wins and evict less adapted ones. So if LLM and AI agents give economics objective advantage (increase productivity and capabilities), than societies that use AI-agents will evict societies that do not use AI-agents. It is undeniable law of nature and no good will or morals can stop it.
Planet itself is also constantly changing. There were multiple catastrophic extinct of species where ~80% species ceased to exist. There were rearrangement of continents, major climate changes much more massive than worst estimations of current global warming and it all happened before human species were born.
So attempting to stop changes is wrong and impossible task in principle, humans and human societies should adapt to changes.
Ancient China is a great example what happens with super morals based societies that refuse technology progress: great thousand old culture have fall to a small army of Europeans with modern military technology (guns etc.). See opium wars. If Western world will refuse to accept technology progress, something like that may occur in opposite direction.
@waddlesplash, let me be direct. Iâve been writing code for 35 years, starting with Z80 assembly. I have seen this exact panic before. When high-level languages like C became popular, hardcore assembly programmers claimed they would âsupplant human intelligenceâ and ruin programming. Then they said the same thing about IDEs and code autocomplete (I personally donât use IDEs or autocomplete, but it would never cross my mind to campaign for banning them).
You ask why people prefer to interact with neural networks instead of asking questions on the forum? The answer is simple: AI doesnât judge. It doesnât tell beginners to âRTFM,â it doesnât mock their ignorance, and it doesnât start ideological debates. If a community wants human connection, it needs to be more patient and welcoming than a machine. Banning tools wonât force people to socialize.
As for AI replacing human intelligence â I actually donât see it as just a dumb calculator. I interact with it as a colleague or a partner (which actually improves the quality of its responses by an order of magnitude). And while I prefer to write code myself, I have long automated many routine processes using local LLMs. In a year or two, this technology will take another massive leap (that is inevitable). We canât just hide from it in our bubble and pretend the elephant isnât in the room. It doesnât replace humans; it expands our capabilities.
You brought up the âpersonal computerâ as a device for personal computing. But if LLMs are just matrix multiplication, then running them is also personal computing, isnât it? For me, âPersonal Computingâ means transferring the means of production from a narrow circle of the elite to every single person.
If one guy, with the help of his AI partner, can write a working UVC webcam driver in a few weeks â something Haiku has lacked for decades â that is not a reason to reproach him for code quality or call him a troll. It is rather a reproach to us that we couldnât manage to do it ourselves, even though we tried.
There is a pithy summary of Norse paganism which goes like this: âThe giants will beat the gods in the end, but I am on the side of the gods.â In other words: âour cause will certainly be defeated; but we still believe in it and shall fight to the last.â Or, for a more general and less religious statement: âA dead thing can go with the stream, but only a living thing can go against it.â
Now, I donât believe in paganism, Norse or otherwise. But I can at least admire them for sticking to their beliefs, even when they were certain they were going to be defeated in the end. They knew what they thought was right, and what ought to be done, and they were going to do it even if they got defeated in the end.
Maybe LLMs will eat the world. Maybe there will be another âmass extinction eventâ, maybe even caused by human folly or arrogance. Maybe, maybe, maybe. But right now, today, neither of those things has happened. And in fact it is still quite possible to fight against those things. And for as long as it is possible to fight against those things, I intend to fight against those things. Maybe the world will change for the better, maybe it will change for the worse: but today, in this moment, if I can make this little corner of the world a little better by defending it from the oncoming tide that will one day drown it, I will accept that victory, as minuscule as it may seem.
If I was responsible for the political future of âthe Western worldâ, I would surely be thinking about all this very differently. But I am not. And, as far as I know, none of us here are. Haiku is not asking anybody to not use LLMs outside Haiku itself; we are (or some of us, anyway) are trying to ask people to not use them for Haiku itself.
I am aware some said they would âruin programmingâ or something to that effect, yes. But did they really claim they would âsupplant human intelligenceâ? That is a different thing entirely; and I am not aware of those claims, really.
The easy availability of calculators means that most people are quite bad at mental math. Skill atrophying is very real. But in fact the person who spends their day doing spreadsheet calculations with pencil and paper is not really using their âhuman intelligenceâ very much. So if we automate the drudgery by using calculators, this potentially frees them up to do less menial work. (Of course they usually just wind up doing other menial work instead, it seemsâŚ)
Automating actually menial, mindless work sounds fine to me, in theory. But in practice, well, as Iâve outlined elsewhere, I want it to be automated by humans formalizing the processes theyâve intuitively created in their head. If we use a machine to âhallucinateâ a process into existence, that doesnât sound good to me.
And of course, thereâs work that seems âdrugerousâ which is in reality not âmindlessâ but requires real thought. This is where a lot of the biggest danger comes from, because people will think âoh, itâs mindless like all these other tasks, Iâll get the LLM to do itâ but in fact itâs not mindless and then we get even worse mistakes from the LLM than usual.
I agree with this. But I think this cuts both ways: if we eliminate all the places that friction happens and people get frustrated, then people wonât grow in patience and âwelcomingnessâ.
This I donât agree with. More and more schools are banning smartphones because theyâre really detrimental to socialization and the social atmosphere in general, despite ostensibly being âjust a toolâ. Banning tools does force people to socialize, at least in spaces where they have no choice to be in.
Perhaps banning these tools will raise the barrier to entry enough that some people wonât socialize at all, and just âbounce offâ instead. Sure, I can admit thatâs likely. But if those people didnât want to socialize in the first place, or rather wanted to socialize only on their terms, well, would they have formed âhuman connectionâ with us anyway?
The elephant surely is in the room. The elephant is doing great damage to the room, as far as I can see. We have had to take various active steps to deal with the damage the elephant is causing, and I suspect we will continue to have to do so. I donât think anyone is advocating for a retreat to some âbubbleâ here. Even a full ban on âAIâ would be equivalent to erecting a new defensive wall that would require regular maintenance, not just âhidingâ.
Mixing personal and ideological choices to refuse to interact with progress makes no sense. Personally, Iâm actively engaged in my life pursuing a whole series of choices that some might consider radical, but this has nothing to do with the tools but with the results. The point isnât what we believe is right, but what is actually useful for achieving a certain result⌠yes, the means justifies the end. According to your enlightened minds, we should just watch the acceleration of all the systems we know, limiting ourselves to boasting about how right and beautiful it is to be a sort of technological hermit⌠but cut it out.
People: please stop making assertions like this that your personal preferences or experiences are definitionally correct. These are subjective issues and subjective values, not objective facts.
I would suggest a less benign example: web search. It encourages you to look for an existing solution instead of thinking about the problem independently.
A lot of tasks are not really about some kind of âprocessâ, but about rote translation from one language to another. Packaging an application is a good example: the repository of an application has a README describing how to build the application, a (for example) pyproject.toml file with a list of dependencies, and a LICENSE file. And the first thing you need to do is to merely translate all that text into a package description.
A language model is a capable translator.
This gets worse as people confuse a text calculator with an intelligence.
Iâm a bit disconcerted by this attitude toward the entire community. Itâs very strange, to say the least.
Treating frustration and friction as a way to âgrow in patienceâ sounds like hazing or a military boot camp. The forum is a place for collaboration and sharing experiences, not for testing peopleâs moral character. If beginners face hostility, they donât learn patience â they just leave.
This is very exclusionary. Many of those who âbounce offâ are talented developers who just want to write software, not play forum politics. By gatekeeping the community like this, we are simply driving away the hands that want to help us.
Waging a doomed battle like a Norse god is romantic, but Haiku is a software project, not a battlefield for a holy war against âgiants.â If we turn our technology into a religious crusade, weâll end up with a very pure, very ethical, but utterly dead and empty operating system. No users, no modern software, and no modern technologies. Completely uncompetitive.
Letâs build an OS for the real world, not a monastery for the chosen few.
I must congratulate everyone on this thread for being quite civil and polite, thats really great to see that we respect everyones opinion. Great values and respect to everyone.
6 months ago I despised AI. Really dispused. 3 months ago tried it for some hobby project. It got me âunstuckâ and moved the project along. And since then Iâve noticed a magnitude (x10) increase in productivity. It hallucinates, gives slop, is wrong 8 times out of 10. But those 2 times it was right it moved my projects along faster than doing the hard slug manually. Its getting better. And for the last 2 months, I use it as advanced search and pair programming. I still review every line, and treat it like a junior. It fumbles and then it doesnt. But at the end, Iâve noticed a magnitude improvement in performance. And you cannot deny the benefits.
Anyhow, I was pretty pig headed and stubborn, but Iâve come to change my views about AI.
Anyhow, its great to see that the community is still respectful, and courtious to all. Thank you.
I guess this point causes some confusion: what it mean ânot to use them for Haiku itselfâ. I guess most of us agree on not use LLM for OS development. But here we are talking about applications, not the OS itself. Where we draw the line? Most of us donât like or use social networks. But what will happen if someone some day create a native Facebook or Twitter client for Haiku? we are to try to ban it too? Or we are going to say to him âwe donât want your aplication hereâ? Donât get me wrong: I am against LLA companies. But I am not against people who uses them to create their apps. We are making a mistake putting everyone in the same bag.
Also, worth to mention that is a bit of inconsistency to be against LLM and host the recipes and patches on a platform (Github) hosted and owned by one of the biggest investors in LLM (Microsoft). But I guess this is a different topic
While the electricity use (and thus ecological footprint) of AI is real, I do hope that the same discussion happened when the masses started Googling, with their screen background set on white even. Same goes for driving your car, using whatever electrical appliance, etc.
What they all have in common is that the first versions are way more inefficient than their most recent versions. The same thing is already happening with AI models. Everyone realizes that the limiting factor will be the energy supply and is starting to optimize. Like 3deyes mentioned as well: Haiku power saving is almost non-existent. Maybe fix that first, before complaining about energy use of other software?
Someone wrote more water usage? To cool data centers? There is no shortage of water on the planet, at all. Only the distribution of (drinkable) water is an issue for some countries and that was already the case even before AI. Desalinization can be done in various ways and itâs available.
Social interaction? A lot of us, myself sometimes included, are probably nerds in some way. A lot of times, I donât even want interaction with people after a long day in the office. Let people choose what they want. Social people can still socialize on irc, FaceTime, or whatever they choose. Others prefer interacting with a LLM.
What works for one, doesnât work for others. Give people freedom and judge them by output. Just like in the real world, in your professional life.
One thing any community must do is decide what to exclude. Whatâs not acceptable within its ranks. And using AI is fundamentally selfish. Itâs anti-community by definition. Heck, by design.
Yâall donât like being judged? Donât do things that impact others. Hint: the only way you can do that is to go live in a cave. Otherwise youâre answerable for your behavior. People will rightly hold you responsible.
This point is not so minor, since companies like OpenAI or Anthropic are the strongest arguments against LLMs. Iâm mostly not guilty of supporting them though, since I use open weight models like Kimi K2 or Qwen.
Some people are very anti big tech (Google, Microsoft, Anthropic, OpenAI, etc) and I understand that if sovereignty or privacy is their issue. Yet, Haiku happily enters the GSoC every year, where the G stands for Google. Nobody complains about that. In the case of AI, you can use models from other companies or countries, or use a stand alone LLM.
The whole open source movement is solving global problems from corporate monopolies, for instance. Each individual which, one day, contribute at his small scale bring his stone to a way bigger project, which make it that today, we, users and developers, have choice to use multiples softwares and softwares platforms, even against the self-interest of some corporate quasi-monopolies. Itâs only âquasiâ thanks to the huge number of small stones contributed by the open source movement since 4/5 decades, maybe.
If to have today what we have in term of open source softwares and platforms one would have to wait that a global organisation with a clear objective was setup around the globe, Iâm sure we will have nothing. It worked because people took the initiative to do it each at their individual scale, and these people communicates between them, not because they were in a global mission all togetherâŚ
