> I used to argue with people on the internet, after about six replies, you realize that you’re speaking to someone incapable of thought
He realises the current woe of things but doesn't realise he's been arguing with bot farms and teams of people hired just for this reason - to sow doom, arguments and engagement.
Around 10 years ago I noticed this happening on trending topics of Twitter - it wasn't that the opponents were stupid because they disagreed, it was that they were simultaneously intelligent and stupid in the way they spoke in a way that I realised I'd never seen in genuine people, making me realise it was probably different people or bots under one account.
If I realised it 10 years ago it was probably happening for at least 15. He wasn't better than these people he was falling into their trap
Haha it’s just a foolish move made by those from the ‘90s. We all internalized a different lesson. The younger generations have learned to brush it off and stop responding. Even if the person is human, when you fail to move toward Aumann Agreement, there is no point in persisting.
Us older ones are still fighting the last war. If only we explain it properly, maybe the bots will understand!
I justify it to myself by saying that even if I'm arguing with a bot, I want to have a counterargument posted for the benefit of the people who are reading the comments
Haha, it’s figurative and requires squinting. “rational actors with common axioms will reach shared conclusions through repeated exchange of information”.
So if you’re not reaching a shared conclusion then you’re not exchanging information, you don’t have shared axioms, or one of you isn’t rational.
If it didn’t land with you, it was just a poor choice of phrase and while I care to attempt to explain it, I don’t care to attempt to defend it.
Values are not different from other positions that people might hold. I believe values and principles are just lossy compressions of positions about specifics because we cannot express them all. When you poke at the edges you find they are fractal. So “what’s your estimate for random variable X?” And “are open borders good?” Are not really different questions so much as questions about different levels of specificity.
Have you heard about the free energy principle and free energy model of emotions? In this model, all brain does is minimize the prediction error, so all values and beliefs are bayesian priors by necessity.
I've now completely stopped engaging with anyone on that platform that I can't reasonably traced back to a real person. If any non-real-name account says anything that remotely feels bad-faith in replies, I block them. Having been on the Internet since the mid 90's when it was the frontier, and everyone helped/trusted everyone else, the policy feels very wrong. But unfortunately Twitter has given no tools or alternatives to deal with the situation.
I have found that the only way to deal with this these days is basically to only use my time to discuss things with people in private, invitation only forums. HN is a rare exception for "I am pretty sure you all are not bots (yet)" and yet anonymous.
Everything else that I care about is in private Signal groups or an equivalent of same with anywhere from 15 to 200 people, and adding any new member means they need to be vouched for by an existing one. And at least 55-75% of the people know each other face to face through industry vertical specific events, trade shows, conferences and similar.
I can't even imagine trying to engage with people on something like x/twitter or other social media. Not knowing whether the 'person' you're replying to is somebody's LLM authoried bot farm.
> And at least 55-75% of the people know each other face to face through industry vertical specific events, trade shows, conferences and similar.
Tbh, my experience with _that_ has been worse than with a random sample of internet background radiation, as people seem to have no idea how to filter correctly and who to vouch for and why.
_Especially_ with people they know via work, conferences, etc. Spaces that do not filter for "genuinely relevant contributer to a conversation" at all, but encourage faking that.
Within this context it's specifically people in the ISP and telecom business, most of whom have been in the business 15, 20 or 25 years now. Some of us started out as junior NOC techs, junior sysadmin/Linux/BSD guys in the late 1990s at different dotcom 1.0 startups.
At a certain scale you know who the people are that have super-admin privileges at very specific ASNs and they are well known in the industry. Basically much as somebody in academia in a journalism school on the west coast would know who is the department head of the school of journalism at a major university in Boston or NY and recognize their name.
Now within a Signal group of 150 people adjacent to the neteng teams at large to medium sized US/Canadian ISPs, it's entirely possible that somebody let an uninvited rando in, but we don't discuss anything super sensitive (that's for more 1:1 messages). A person who has been lurking for weeks or months and suddenly starts chatting in an inauthentic way would be found out very quickly.
One of the things that ISP technical operations groups do very well is filter for non-authentic participant in something, because sales persons getting into groups of technical discussion with the angle of selling some new thing is a perpetual problem and is a well known factor to deal with.
Also sales people generally don't do well when a complex operational-related technical matter is put in front of them in a fast paced environment where they can't engage their engineering team to get an accurate reply. Ask anyone that's ever tried to get their direct cellphone number removed from a Cogent IP transit sales person's CRM system....
Yes, I got that, but what I actually mean is that these criteria do not necessarily lead to good conversations.
I know the kind of space you are referring to. I'm in those too.
But my point is that they regularly disappoint me.
Which is usually worse disappointment than with randoms on the internet, because the people should know better but don't.
With these tight knit spaces you get all the social dysfunctions of people liking each other not for merit but because they've been around each other for long, and that's - for me personally - worse than the public alternative, as you suddenly do not argue with logic, but with the tribe and social cohesion.
That feels like the intellectual equivalent of limiting your career opportunities within a small tiny villages vs. a large city. There is so much more in the city.
It's not that I don't read a great deal from a wide variety of sources, but that I reserve the small amount of free time I have per day for back-and-forth discussions (like this one) with groups of people I am fairly sure are not sock puppets. The amount of oversight that dang exercises on HN to ensure that frequent and verbose commenters are authentic people should not go without notice.
I have a theory that a good part of the filtering for authentic humans has already been done here, which gives me a higher level of confidence that I'm not metaphorically shouting into the void by conversing with somebody's LLM or social media influence apparatus.
I have moved back to my small hometown, from a large city.
My mental health improved significantly. I actually feel valuable at work. I feel pride in contributing to my local community. And the paycheck isn't actually worse.
> discuss things with people in private, invitation only forums
One of the great things about not having ads on rangerovers.pub is that Google is just not interested in indexing it, because it doesn't make them any money. It shows up here and there in search results, sporadically, especially if you know a username or phrase specific to the site, but other than that if you want to know about it you need to be told about it.
I was never a fan of web forums but they are a nice compromise between really oldschool mailing lists and FAANG content-to-eyeball transformers.
I have seen a bit of a resurgence in your classic 2005-era type interface php bulletin board in a browser, as people relegate their facebook/instagram/whatever accounts to only keeping up with family stuff posted by boomers who won't use anything else, or things like niche neighborhood based buy & sell groups (or community notice groups in like a town of 12,000).
I guess there's at least 3 or more active and currently developed software packages with open source licenses, the flaskbb that site uses, simplemachinesforum, the classic phpbb, probably some others I've forgotten about... The LAMP stack dependencies for all seem very mundane and very lightweight.
This is the only way to realistically deal with it but it’s not scalable. A platform is only as useful as its non-bot users. On Twitter even the positive replies are likely bots. Real engagement is limited for most users.
On a site like HN, blocklists are very effective at keeping people off your feed.
Same. (I would love to say, but they still get me way too often)
I can also highly encourage people to keep notes and do some basic OSINT. The effective internet is smaller than one might think, so that proves useful time and time again.
If something seems _odd_, I suggest looking up _why_ it might be odd.
Gathering context, essentially.
Like "Okay, this guy is weird. Aah, okay, LinkedIn says that he works there. Okay _now_ that makes sense".
Based on that, you can then decide how to approach the (previously failing) interaction.
__
HN is a very easy place for that, because, to give somewhat concrete examples, if someone is shilling for something and seems unreachable for common sense, LinkedIn usually tells you that their salary depends on that.
And people play very open here, because this is treated as a business networking event.
>>anyone on that platform that I can't reasonably traced back to a real person
Not sure why that makes any difference - I've had plenty of arguments on Facebook with people who are perfectly happy to spew racist and/or conspiratorial bullshit while having their full names, their holiday photos and photos of their children and/or grandchildren attached to their identity. The weird thing is that these people are 99/100 times incapable of actually having an argument, they either start insulting you straight away or say some variation of "if you don't like what I'm saying then leave" or actually majority of time "what does a foreigner like you know about this topic".
The real names policy has yielded zero of the promised effects imho.
>it wasn't that the opponents were stupid because they disagreed, it was that they were simultaneously intelligent and stupid in the way they spoke in a way that I realised I'd never seen in genuine people, making me realise it was probably different people or bots under one account.
I believed they were all bots until I actually interacted with such a person in real life. It's terrifying.
I used to think that the people who could not apply any semblance of logical reasoning online were just bots or trolls, but then I started meeting them in real life.
One of the first things I tell the junior/mid-level developers I mentor is "You can't debug something just by reading the code." We all have a mental model of how our code works, and it's usually a bit wrong. Bugs are the real world manifestations of those mistakes. When you read the code it's all filtered through your model, and that makes you blind to seeing why something unexpected happened. In order to debug something you have to be able to put the system in the state where the bug happens to see why it occurred.
LLMs generally only debug systems by reading the code with whatever information you give them in a prompt. The image in the article is meta-prompt - the prompt is whatever comes from the vision model the AI happens to use to 'understand' the red circle annotation. That won't work. To successfully debug what's going on it will need much better state information. Has the 'shelf' been explained to is? Is the contrast and lack of shadows in the image messing up the vision model? Why isn't the 'lid' in the image? And so on.
LLMs are clever but they're not magical. Treat them like a naive junior dev. Give them enough data about the state of something to understand it properly.
Hmmm, I don't think that's necessarily true. Often times once I have witnessed a bug, I have found it just by reading through the code with the behaviour of the bug in mind.
For LLMs, this is likely to be disproportionately effective as well: especially because they don't really build up a persistent view of the codebase, they're generally re-reading it each session, and they tend to be surprisingly good at predicting the behaviour of code.
(That said, knowing where and how to gather more evidence to make things clearer is a pretty core skill in troubleshooting, so it's generally good advice anyhow)
I call this a tautological mental model. You can read the code over and over again, but your second reading will be mostly an echo of the mental model you built up in your first.
I completely understand this. I’ve worked on robot hands and 5.6/Fable 5 were practically useless at helping me debug anything visually.
What I have found super useful actually is having models make a interactive 3d viewer in which I use move / highlight / paint (soft body painting directly onto the geometry for issues and different colors mean different failures). This gives a much better way to communicate the physical relationships and positions that are hard to get across in a labeled screenshot.
Its for sure still a lot of manual work so the "seeing-eye dog" description definitely holds. But I have found that after a couple of examples with the extra context the model gets much better at handling the problem and becomes useful.
A bit off topic, but I absolutely love the little robot on the author's main project's landing page (https://rerun.io/).
I normally condemn mouse hijacking, but this implementation will be allowed.
Rerun is pretty dope, but I'm not sure how you came to the conclusion that it's the "author's main project"? There is a whole company backing it, with no affiliation that I could find, apart from the author being an occasional contributor?
They mentioned "... and my visualizer tool comes with an MCP server ...", with a link to the rerun project. I guess it makes more sense that his visualizer tool uses rerun...Sorry for the confusion from my side.
I've found that LLMs are specifically bad at a certain kind of debugging, though I can't quite put my finger on what that is.
"Spot the bug in this code" when the code can be looked at and pattern-matched against bugginess is something they seem really good at.
Some parts of debugging, like "Here is this logfile, what do you think is going on?" are also surprisingly good.
It's that thing kind of in the middle -- I know it when I see it honestly is the best way I can put it into words. An example from recently, I'm receiving some bad data on a network message parser. Immediately I don't know whether it's a my-side or their-side thing, but I know if I try and just vaguely describe the behaviour to the LLM it will start churning tokens.
My current approach to problems like this is -- I need to tell the LLM what it needs to do to give itself the data it needs to solve the problem. My first reaction now isn't "It's not working, there's a bug, it's not doing X". It's "Okay, this isn't quite working properly; I need you to add some debug logging around X, Y and Z so we can figure this out". That tends to avoid spirals and get me out of the situation much more quickly.
The seeing eye dog analogy is pretty apt actually. I would love to see some transcripts from the author if they are able.
Edit to add: I think the 'thing' I'm alluding to might be -- if I have trouble expressing the buggy behaviour clearly in words, then I know it's probably going to be a fair few back-and-forths with the LLM to get something; the harder I find it to concisely describe, the more risk that it'll fall into a pit. Doubly so if I offer up a hypothesis which turns out to be wrong.
LM still can not see and understand the desktop app UI on a level that is acceptable for testing.
All the advances in coding are from web dev and thanks to the nature of html UIs.
Try to develop a desktop app and it’s like working with a legally blind person who can see some part of the screen is they squint in a certain way but surely will miss all minor details.
> I often handwrite the code myself, but I’ve found that LLM coding assistants’ limitless patience ameliorates the drudgiest work of coding.
I've been using a few different "smart" LLM to work on an analysis, parsing, search and correlation tool that ultimately deals with a 5.5GB on disk (with indexes) mariadb database that has its origin as a federal government department's 905,000 row plain text CSV file.
There are a ridiculous number of data entry errors and just plain weird fuckups in the data origin that don't seem they will be ameliorated any time soon, so automating the drudge work of cleaning it up and rectifying it into something usable is a textbook case for this. Very pleased with the results so far.
I mean there's a reason why we're doing MoCap for video games.
If computers were good at this, we wouldn't be needing that. But actual motion and all seems to be much more complex than the systems can predict, apparently.
Also.. uh.. isn't this.. good? I thought AI was to steal all our jobs.
___
Beside that, kinda weird self-description.
Isn't the computer executing your commands and you're just filling in where it cannot do that?
Being that dog implies that the computer is in the driver seat.
I mean it's supposed to be a joke I guess, but I read it as one that leaks internal metadata which seems to be incorrectly calibrated.
It is weird if you think about it this way: it is AI that waits for you, its ‘eyes’, to provide a feedback so it can continue working. It literally uses you as its organ.
<!!spoiler ahead!!>There is a TV show called Person of Interest <!!spoiler ahead!!>, where Machine (AI connected to Internet and CCTV networks) has no legs or eyes, so when it needs to go and check something not covered by CCTV feeds, it gives instructions to a real person. In the show it is called an ‘analog interface.’
If intern is making a breakfast, and boss is the one who suddenly runs to the grocery store because eggs are missing, is boss still the one in a driver’s seat?
From the original goal point of view yes, as it was boss who has initiated whole breakfast procedure. But from an execution standpoint it is intern who gives its boss a job of a grocery store run. He could give same job to anyone else, boss as a persona is irrelevant here.
>Also.. uh.. isn't this.. good? I thought AI was to steal all our jobs.
the problem is, the CEOs still think AI solves their problems ie minimize paid jobs
> All that’s left is the dumbest workflow possible: I fire up the debug viewer myself, look around for weird mistakes, then take a screenshot and tell the language model how badly it messed up this time. Eventually I just decided to do all the debugging work myself, so I would at least get to do the fun part too.
I call this "thanoscoding" for two reasons:
1. "Fine, I'll do it myself"
2. In the past I found I have to "snap away" the mess the LLM made in order to start afresh from a known good state (generally with git reset). But that was 1-2 generations ago when it comes to models. GPT6 Astra probably does things right the first time, 90% of the time.
I like to think of training and improving AIs as bringing freedom to the world. The useless toil and labor associated with rebuilding the same solutions into different contexts is finally at an end.
Coding was never the reward. Acting as a translator for a machine is far worse than allowing the machine to solve the mundane parts and leave you with bigger building blocks to play with.
After 25 years I have to confess I hated being a software engineer. It felt like grinding in a video game.
Now I can finally create and innovate at the speed of thought, and I'm very grateful to have this technology now.
For me code still is the reward; I made and make my fortunes with boring code people on HN say no one needs or wants, my hobby is, and has been for 45 years, writing, perfecting and optimizing code manually until I find it perfect. I have been working for 10 years on a programming language and OS (with niche 2 dbs that are now prod quality and we use) and those are the best moment of my day, I couldn’t care less what anyone thinks of it or ever uses it. But the lessons learned do flow into LLMs to write the boring code and I can say that our million$ paying LoB code can handle 100s req/s on shite hardware and even if the LLM did a crap job. Which is almost 100s times more than the client will ever need.
Giant corpos hold every sliver of your so called freedom, you dont innovate, you repeat what others created before you. You use a tool that shackles your thoughts and creativity in a never before seen way, what you perceive is an illusion of liberty that is no present. Without others you are nothing and they can take it away any seconds, you are an addict, not an innovator.
I can see where you're coming from, and I share the sentiment and resentment to a large degree, but I think you might be not doing reality justice.
It is true that closed weights models are a big issue. It is also true that LLM-generated solutions usually drift towards a median.
But that is not the dead end you think it might be.
Most coding work is repetitive boilerplate, and most typing is just.. well.. typing. Miserable work I too did not really enjoy. What I did enjoy were the end results, and that was just a necessary step to get there.
Now that's less the case than it was before we had LLMs, and for that I too am glad.
___
I think the article headline might've primed you (and me, fwiw) to reading the comment you're replying to as passive. But if you just look at the words of it, that might not actually be the case.
Your point is true to some extend, I personally host my own "cloud" storage..it comes with drawbacks, same is true for email..especially email, social media is a society constructs, by definition its reliant on others but many will argue that its not worth keeping anyway (although I wonder how one stays human in this world then, the meat space seems with so many barrieres in this age).
I feel very much like a luddite..or some artisan of a bygone age. The guy above laments how he never enjoyed coding, this is an interesting sentiment, I know quite a few people who enjoy the craft itself, myself included. The ability to form words that have meaning, that create something from nothing. Sure, the words are just a tool, but its something _I_ can master, not some abstract wish machine that may change its functionality tommorow. Obviously one can argue that the computer itself is in this case the one that creates and not me, its not magic that just works with me, but the machine I got obeys me and me alone..another reason why personal computing is important.
Serious question - what do you do for fun? I find fishing with friends enjoyable, and using my hands tidying up the old place I bought. It’s not innovating software I’ll never use but I’ll never tire of writing the same old ASP that delivers my clients the results they’re after.
> I like to think of training and improving AIs as bringing freedom to the world.
Yes, when I think of OpenAI and Anthrophic and Google and Meta and any AI labs and their intentions, i cry a single tear for how these great instituitions are working so hard to bring freedom for humanity.
AI owned by a few companies is more likely to put the majority right back to serfdom. You're a privileged fool to believe that freedom is the likely outcome of the current stampede.
With a username that appears to cheer for a sociopath¹, I doubt reason will convince you.
It was obviously a sarcastic joke. I get it if you didn't catch that, but do you think my username glorifies Sam Altman? And is that a reason to talk negatively about me? You doubt reason will convince me? And at the end, you call *me* a privileged fool. Wow, just wow.
I hope you don't consider NWA a racial hate group given their name.
Also, I don't think you can cite a subjective take as objectively correct by including it as a footnote. Why do you need to medically diagnose him to say what he's doing is fucked up?
Why does he have to be literally the devil for outrage to work?
He could also not be a sociopath, but still his actions will be his actions, and it is okay to criticize them. Next what, you're going to link to an article that says Sam Altman is also a homosexual, and we must be morally outraged at that next? Because that is where your direction is going, instead of arguing about the ill effects of AI.
I share the sentiment but the question here is how long can we maintain the balance point where the human in the agentic loop is required. It might be a window lasting only a few years, or for the foreseeable future; I think the answer lies the opaque compute economics of the frontier lab: how well models keep scaling and how economically sustainable is serving those models under the current market conditions.
If my job gets automated, I'll find something else to do. I wouldn't have wanted lamplighters to succeed in preventing electrification, so it would be unfair for me to prevent the automation of my job if it can be done.
I use Astra to drive Fable; I drivel into my phone while walking in the forest and it builds. I don’t need to check; I do as clients need to pay, but it always is great. And it surprises me with things I did not know were possible even (never encountered them before so why would I know). We are at the point where our clients send voice messages and they get what they want without humans basically. This costs 10+10 max2 subs but that’s nothing compared to hiring people. We didn’t fire anyone; we just have 100+ more clients and make almost 50x more money. It’s boring but great as long as it lasts, we are already where you say for what enterprises generally need for the boring parts. That’s 99%.
I find I still add value, but I don't know if my value is real. Am I just biased and expect things to be done a certain way and penalize the model for doing something different? and am I providing the model with enough high quality context to align with my expectations in one-shot?
What about cases where a bug is only obvious in motion? Or where you're working with a SBC and the bug is that you forgot a cable? Screenshots alone aren't enough. The authors point stands.
> I used to argue with people on the internet, after about six replies, you realize that you’re speaking to someone incapable of thought
He realises the current woe of things but doesn't realise he's been arguing with bot farms and teams of people hired just for this reason - to sow doom, arguments and engagement.
Around 10 years ago I noticed this happening on trending topics of Twitter - it wasn't that the opponents were stupid because they disagreed, it was that they were simultaneously intelligent and stupid in the way they spoke in a way that I realised I'd never seen in genuine people, making me realise it was probably different people or bots under one account.
If I realised it 10 years ago it was probably happening for at least 15. He wasn't better than these people he was falling into their trap
Haha it’s just a foolish move made by those from the ‘90s. We all internalized a different lesson. The younger generations have learned to brush it off and stop responding. Even if the person is human, when you fail to move toward Aumann Agreement, there is no point in persisting.
Us older ones are still fighting the last war. If only we explain it properly, maybe the bots will understand!
I justify it to myself by saying that even if I'm arguing with a bot, I want to have a counterargument posted for the benefit of the people who are reading the comments
> If only we explain it properly, maybe the bots will understand!
I'm going to be thinking about this perfect summary of my frustration of the last 20 years all day
Arguing on the internet is not about convincing the other side. It's about winning the majority opinion and getting more votes. Almost like politics.
Obvioulsy not true for all the arguments, only a subset of those not worth having.
I googled Aumann Agreement and it’s all about agreeing about whether things are true/false/likely/unlikely.
I don’t understand your point at all in this context. What’s “Aumann Agreement” in the context of a discussion about values?
Haha, it’s figurative and requires squinting. “rational actors with common axioms will reach shared conclusions through repeated exchange of information”.
So if you’re not reaching a shared conclusion then you’re not exchanging information, you don’t have shared axioms, or one of you isn’t rational.
If it didn’t land with you, it was just a poor choice of phrase and while I care to attempt to explain it, I don’t care to attempt to defend it.
Values are not different from other positions that people might hold. I believe values and principles are just lossy compressions of positions about specifics because we cannot express them all. When you poke at the edges you find they are fractal. So “what’s your estimate for random variable X?” And “are open borders good?” Are not really different questions so much as questions about different levels of specificity.
My incomplete self-note about this here https://wiki.roshangeorge.dev/w/Principles_Are_Compressed_Im...
Have you heard about the free energy principle and free energy model of emotions? In this model, all brain does is minimize the prediction error, so all values and beliefs are bayesian priors by necessity.
Appreciate the writeup, thanks.
The outcomes those values achieve.
I've now completely stopped engaging with anyone on that platform that I can't reasonably traced back to a real person. If any non-real-name account says anything that remotely feels bad-faith in replies, I block them. Having been on the Internet since the mid 90's when it was the frontier, and everyone helped/trusted everyone else, the policy feels very wrong. But unfortunately Twitter has given no tools or alternatives to deal with the situation.
I have found that the only way to deal with this these days is basically to only use my time to discuss things with people in private, invitation only forums. HN is a rare exception for "I am pretty sure you all are not bots (yet)" and yet anonymous.
Everything else that I care about is in private Signal groups or an equivalent of same with anywhere from 15 to 200 people, and adding any new member means they need to be vouched for by an existing one. And at least 55-75% of the people know each other face to face through industry vertical specific events, trade shows, conferences and similar.
I can't even imagine trying to engage with people on something like x/twitter or other social media. Not knowing whether the 'person' you're replying to is somebody's LLM authoried bot farm.
> And at least 55-75% of the people know each other face to face through industry vertical specific events, trade shows, conferences and similar.
Tbh, my experience with _that_ has been worse than with a random sample of internet background radiation, as people seem to have no idea how to filter correctly and who to vouch for and why.
_Especially_ with people they know via work, conferences, etc. Spaces that do not filter for "genuinely relevant contributer to a conversation" at all, but encourage faking that.
Within this context it's specifically people in the ISP and telecom business, most of whom have been in the business 15, 20 or 25 years now. Some of us started out as junior NOC techs, junior sysadmin/Linux/BSD guys in the late 1990s at different dotcom 1.0 startups.
At a certain scale you know who the people are that have super-admin privileges at very specific ASNs and they are well known in the industry. Basically much as somebody in academia in a journalism school on the west coast would know who is the department head of the school of journalism at a major university in Boston or NY and recognize their name.
Now within a Signal group of 150 people adjacent to the neteng teams at large to medium sized US/Canadian ISPs, it's entirely possible that somebody let an uninvited rando in, but we don't discuss anything super sensitive (that's for more 1:1 messages). A person who has been lurking for weeks or months and suddenly starts chatting in an inauthentic way would be found out very quickly.
One of the things that ISP technical operations groups do very well is filter for non-authentic participant in something, because sales persons getting into groups of technical discussion with the angle of selling some new thing is a perpetual problem and is a well known factor to deal with.
Also sales people generally don't do well when a complex operational-related technical matter is put in front of them in a fast paced environment where they can't engage their engineering team to get an accurate reply. Ask anyone that's ever tried to get their direct cellphone number removed from a Cogent IP transit sales person's CRM system....
Yes, I got that, but what I actually mean is that these criteria do not necessarily lead to good conversations.
I know the kind of space you are referring to. I'm in those too. But my point is that they regularly disappoint me.
Which is usually worse disappointment than with randoms on the internet, because the people should know better but don't.
With these tight knit spaces you get all the social dysfunctions of people liking each other not for merit but because they've been around each other for long, and that's - for me personally - worse than the public alternative, as you suddenly do not argue with logic, but with the tribe and social cohesion.
That feels like the intellectual equivalent of limiting your career opportunities within a small tiny villages vs. a large city. There is so much more in the city.
It's not that I don't read a great deal from a wide variety of sources, but that I reserve the small amount of free time I have per day for back-and-forth discussions (like this one) with groups of people I am fairly sure are not sock puppets. The amount of oversight that dang exercises on HN to ensure that frequent and verbose commenters are authentic people should not go without notice.
I have a theory that a good part of the filtering for authentic humans has already been done here, which gives me a higher level of confidence that I'm not metaphorically shouting into the void by conversing with somebody's LLM or social media influence apparatus.
I have moved back to my small hometown, from a large city.
My mental health improved significantly. I actually feel valuable at work. I feel pride in contributing to my local community. And the paycheck isn't actually worse.
Sounds just like gambling, honestly.
Which might be a winning strategy, but also might not be.
> discuss things with people in private, invitation only forums
One of the great things about not having ads on rangerovers.pub is that Google is just not interested in indexing it, because it doesn't make them any money. It shows up here and there in search results, sporadically, especially if you know a username or phrase specific to the site, but other than that if you want to know about it you need to be told about it.
I was never a fan of web forums but they are a nice compromise between really oldschool mailing lists and FAANG content-to-eyeball transformers.
I have seen a bit of a resurgence in your classic 2005-era type interface php bulletin board in a browser, as people relegate their facebook/instagram/whatever accounts to only keeping up with family stuff posted by boomers who won't use anything else, or things like niche neighborhood based buy & sell groups (or community notice groups in like a town of 12,000).
I guess there's at least 3 or more active and currently developed software packages with open source licenses, the flaskbb that site uses, simplemachinesforum, the classic phpbb, probably some others I've forgotten about... The LAMP stack dependencies for all seem very mundane and very lightweight.
This is the only way to realistically deal with it but it’s not scalable. A platform is only as useful as its non-bot users. On Twitter even the positive replies are likely bots. Real engagement is limited for most users.
On a site like HN, blocklists are very effective at keeping people off your feed.
Same. (I would love to say, but they still get me way too often)
I can also highly encourage people to keep notes and do some basic OSINT. The effective internet is smaller than one might think, so that proves useful time and time again.
Can you elaborate on the OSINT suggestion? Do you mean on the people you engage with?
Yes, exactly. Though, not by default.
If something seems _odd_, I suggest looking up _why_ it might be odd. Gathering context, essentially.
Like "Okay, this guy is weird. Aah, okay, LinkedIn says that he works there. Okay _now_ that makes sense".
Based on that, you can then decide how to approach the (previously failing) interaction.
__
HN is a very easy place for that, because, to give somewhat concrete examples, if someone is shilling for something and seems unreachable for common sense, LinkedIn usually tells you that their salary depends on that.
And people play very open here, because this is treated as a business networking event.
>>anyone on that platform that I can't reasonably traced back to a real person
Not sure why that makes any difference - I've had plenty of arguments on Facebook with people who are perfectly happy to spew racist and/or conspiratorial bullshit while having their full names, their holiday photos and photos of their children and/or grandchildren attached to their identity. The weird thing is that these people are 99/100 times incapable of actually having an argument, they either start insulting you straight away or say some variation of "if you don't like what I'm saying then leave" or actually majority of time "what does a foreigner like you know about this topic".
The real names policy has yielded zero of the promised effects imho.
>it wasn't that the opponents were stupid because they disagreed, it was that they were simultaneously intelligent and stupid in the way they spoke in a way that I realised I'd never seen in genuine people, making me realise it was probably different people or bots under one account.
I believed they were all bots until I actually interacted with such a person in real life. It's terrifying.
I used to think that the people who could not apply any semblance of logical reasoning online were just bots or trolls, but then I started meeting them in real life.
Sure, they exist, but online they are simulatenously unable to apply any logic and then surprisingly coherent and intelligent in the next reply.
I'd love to know if there's a documentary or something that has been done about this, that I could point non-techie people in my life to.
This is some schizo thinking. People you interact with online almost certainly aren't bots or paid.
Certainly aren't bots is dubious, although if there's actual interactions it's likely they aren't bots.
Paid, lmao I wish. Nobody gets paid to repeat propaganda, generally.
One of the first things I tell the junior/mid-level developers I mentor is "You can't debug something just by reading the code." We all have a mental model of how our code works, and it's usually a bit wrong. Bugs are the real world manifestations of those mistakes. When you read the code it's all filtered through your model, and that makes you blind to seeing why something unexpected happened. In order to debug something you have to be able to put the system in the state where the bug happens to see why it occurred.
LLMs generally only debug systems by reading the code with whatever information you give them in a prompt. The image in the article is meta-prompt - the prompt is whatever comes from the vision model the AI happens to use to 'understand' the red circle annotation. That won't work. To successfully debug what's going on it will need much better state information. Has the 'shelf' been explained to is? Is the contrast and lack of shadows in the image messing up the vision model? Why isn't the 'lid' in the image? And so on.
LLMs are clever but they're not magical. Treat them like a naive junior dev. Give them enough data about the state of something to understand it properly.
Hmmm, I don't think that's necessarily true. Often times once I have witnessed a bug, I have found it just by reading through the code with the behaviour of the bug in mind.
For LLMs, this is likely to be disproportionately effective as well: especially because they don't really build up a persistent view of the codebase, they're generally re-reading it each session, and they tend to be surprisingly good at predicting the behaviour of code.
(That said, knowing where and how to gather more evidence to make things clearer is a pretty core skill in troubleshooting, so it's generally good advice anyhow)
Also Claude Code is very good at writing small scripts/on-off test cases to confirm bugs, so I wouldn't even say the initial premise is correct.
I call this a tautological mental model. You can read the code over and over again, but your second reading will be mostly an echo of the mental model you built up in your first.
I completely understand this. I’ve worked on robot hands and 5.6/Fable 5 were practically useless at helping me debug anything visually.
What I have found super useful actually is having models make a interactive 3d viewer in which I use move / highlight / paint (soft body painting directly onto the geometry for issues and different colors mean different failures). This gives a much better way to communicate the physical relationships and positions that are hard to get across in a labeled screenshot.
Its for sure still a lot of manual work so the "seeing-eye dog" description definitely holds. But I have found that after a couple of examples with the extra context the model gets much better at handling the problem and becomes useful.
A bit off topic, but I absolutely love the little robot on the author's main project's landing page (https://rerun.io/). I normally condemn mouse hijacking, but this implementation will be allowed.
Rerun is pretty dope, but I'm not sure how you came to the conclusion that it's the "author's main project"? There is a whole company backing it, with no affiliation that I could find, apart from the author being an occasional contributor?
They mentioned "... and my visualizer tool comes with an MCP server ...", with a link to the rerun project. I guess it makes more sense that his visualizer tool uses rerun...Sorry for the confusion from my side.
I've found that LLMs are specifically bad at a certain kind of debugging, though I can't quite put my finger on what that is.
"Spot the bug in this code" when the code can be looked at and pattern-matched against bugginess is something they seem really good at.
Some parts of debugging, like "Here is this logfile, what do you think is going on?" are also surprisingly good.
It's that thing kind of in the middle -- I know it when I see it honestly is the best way I can put it into words. An example from recently, I'm receiving some bad data on a network message parser. Immediately I don't know whether it's a my-side or their-side thing, but I know if I try and just vaguely describe the behaviour to the LLM it will start churning tokens.
My current approach to problems like this is -- I need to tell the LLM what it needs to do to give itself the data it needs to solve the problem. My first reaction now isn't "It's not working, there's a bug, it's not doing X". It's "Okay, this isn't quite working properly; I need you to add some debug logging around X, Y and Z so we can figure this out". That tends to avoid spirals and get me out of the situation much more quickly.
The seeing eye dog analogy is pretty apt actually. I would love to see some transcripts from the author if they are able.
Edit to add: I think the 'thing' I'm alluding to might be -- if I have trouble expressing the buggy behaviour clearly in words, then I know it's probably going to be a fair few back-and-forths with the LLM to get something; the harder I find it to concisely describe, the more risk that it'll fall into a pit. Doubly so if I offer up a hypothesis which turns out to be wrong.
Those three colored shapes at the bottom look a lot like the EPA logo, a former grocery store chain: https://de.wikipedia.org/wiki/EPA_%28Warenhaus%29?wprov=sfla...
LM still can not see and understand the desktop app UI on a level that is acceptable for testing. All the advances in coding are from web dev and thanks to the nature of html UIs. Try to develop a desktop app and it’s like working with a legally blind person who can see some part of the screen is they squint in a certain way but surely will miss all minor details.
You could setup a webcam and have a vision llm stalk the breakroom of left over pizza and alert you.
> I often handwrite the code myself, but I’ve found that LLM coding assistants’ limitless patience ameliorates the drudgiest work of coding.
I've been using a few different "smart" LLM to work on an analysis, parsing, search and correlation tool that ultimately deals with a 5.5GB on disk (with indexes) mariadb database that has its origin as a federal government department's 905,000 row plain text CSV file.
There are a ridiculous number of data entry errors and just plain weird fuckups in the data origin that don't seem they will be ameliorated any time soon, so automating the drudge work of cleaning it up and rectifying it into something usable is a textbook case for this. Very pleased with the results so far.
I mean there's a reason why we're doing MoCap for video games. If computers were good at this, we wouldn't be needing that. But actual motion and all seems to be much more complex than the systems can predict, apparently.
Also.. uh.. isn't this.. good? I thought AI was to steal all our jobs.
___
Beside that, kinda weird self-description.
Isn't the computer executing your commands and you're just filling in where it cannot do that?
Being that dog implies that the computer is in the driver seat.
I mean it's supposed to be a joke I guess, but I read it as one that leaks internal metadata which seems to be incorrectly calibrated.
<!!spoiler ahead!!>There is a TV show called Person of Interest <!!spoiler ahead!!>, where Machine (AI connected to Internet and CCTV networks) has no legs or eyes, so when it needs to go and check something not covered by CCTV feeds, it gives instructions to a real person. In the show it is called an ‘analog interface.’
> It literally uses you as its organ.
But it isn't. That's my point.
I told the clanker "hey do that", and like the intern/junior it emulates, it eventually says "boss! Help! I can't do this alone".
It is I who is in the driver seat.
If intern is making a breakfast, and boss is the one who suddenly runs to the grocery store because eggs are missing, is boss still the one in a driver’s seat?
From the original goal point of view yes, as it was boss who has initiated whole breakfast procedure. But from an execution standpoint it is intern who gives its boss a job of a grocery store run. He could give same job to anyone else, boss as a persona is irrelevant here.
>Also.. uh.. isn't this.. good? I thought AI was to steal all our jobs. the problem is, the CEOs still think AI solves their problems ie minimize paid jobs
> All that’s left is the dumbest workflow possible: I fire up the debug viewer myself, look around for weird mistakes, then take a screenshot and tell the language model how badly it messed up this time. Eventually I just decided to do all the debugging work myself, so I would at least get to do the fun part too.
I call this "thanoscoding" for two reasons:
1. "Fine, I'll do it myself"
2. In the past I found I have to "snap away" the mess the LLM made in order to start afresh from a known good state (generally with git reset). But that was 1-2 generations ago when it comes to models. GPT6 Astra probably does things right the first time, 90% of the time.
I like to think of training and improving AIs as bringing freedom to the world. The useless toil and labor associated with rebuilding the same solutions into different contexts is finally at an end.
Coding was never the reward. Acting as a translator for a machine is far worse than allowing the machine to solve the mundane parts and leave you with bigger building blocks to play with.
After 25 years I have to confess I hated being a software engineer. It felt like grinding in a video game.
Now I can finally create and innovate at the speed of thought, and I'm very grateful to have this technology now.
For me code still is the reward; I made and make my fortunes with boring code people on HN say no one needs or wants, my hobby is, and has been for 45 years, writing, perfecting and optimizing code manually until I find it perfect. I have been working for 10 years on a programming language and OS (with niche 2 dbs that are now prod quality and we use) and those are the best moment of my day, I couldn’t care less what anyone thinks of it or ever uses it. But the lessons learned do flow into LLMs to write the boring code and I can say that our million$ paying LoB code can handle 100s req/s on shite hardware and even if the LLM did a crap job. Which is almost 100s times more than the client will ever need.
You got nothing.
Giant corpos hold every sliver of your so called freedom, you dont innovate, you repeat what others created before you. You use a tool that shackles your thoughts and creativity in a never before seen way, what you perceive is an illusion of liberty that is no present. Without others you are nothing and they can take it away any seconds, you are an addict, not an innovator.
I can see where you're coming from, and I share the sentiment and resentment to a large degree, but I think you might be not doing reality justice.
It is true that closed weights models are a big issue. It is also true that LLM-generated solutions usually drift towards a median.
But that is not the dead end you think it might be.
Most coding work is repetitive boilerplate, and most typing is just.. well.. typing. Miserable work I too did not really enjoy. What I did enjoy were the end results, and that was just a necessary step to get there.
Now that's less the case than it was before we had LLMs, and for that I too am glad.
___
I think the article headline might've primed you (and me, fwiw) to reading the comment you're replying to as passive. But if you just look at the words of it, that might not actually be the case.
The same can be said of stuff like cloud storage and email and social media. And yet here we are.
Your point is true to some extend, I personally host my own "cloud" storage..it comes with drawbacks, same is true for email..especially email, social media is a society constructs, by definition its reliant on others but many will argue that its not worth keeping anyway (although I wonder how one stays human in this world then, the meat space seems with so many barrieres in this age).
I feel very much like a luddite..or some artisan of a bygone age. The guy above laments how he never enjoyed coding, this is an interesting sentiment, I know quite a few people who enjoy the craft itself, myself included. The ability to form words that have meaning, that create something from nothing. Sure, the words are just a tool, but its something _I_ can master, not some abstract wish machine that may change its functionality tommorow. Obviously one can argue that the computer itself is in this case the one that creates and not me, its not magic that just works with me, but the machine I got obeys me and me alone..another reason why personal computing is important.
You lost leverage. Now be great full for the crumbs. If there remain any.
Viva la revolution!
Serious question - what do you do for fun? I find fishing with friends enjoyable, and using my hands tidying up the old place I bought. It’s not innovating software I’ll never use but I’ll never tire of writing the same old ASP that delivers my clients the results they’re after.
> I like to think of training and improving AIs as bringing freedom to the world.
Yes, when I think of OpenAI and Anthrophic and Google and Meta and any AI labs and their intentions, i cry a single tear for how these great instituitions are working so hard to bring freedom for humanity.
Freedom is slavery!
AI owned by a few companies is more likely to put the majority right back to serfdom. You're a privileged fool to believe that freedom is the likely outcome of the current stampede.
With a username that appears to cheer for a sociopath¹, I doubt reason will convince you.
¹ https://futurism.com/artificial-intelligence/sources-sam-alt...
Their /s was implied
(Like, Meta’s intentions? lol)
I hope so.
It was obviously a sarcastic joke. I get it if you didn't catch that, but do you think my username glorifies Sam Altman? And is that a reason to talk negatively about me? You doubt reason will convince me? And at the end, you call *me* a privileged fool. Wow, just wow.
I hope you don't consider NWA a racial hate group given their name.
Also, I don't think you can cite a subjective take as objectively correct by including it as a footnote. Why do you need to medically diagnose him to say what he's doing is fucked up?
Why does he have to be literally the devil for outrage to work?
He could also not be a sociopath, but still his actions will be his actions, and it is okay to criticize them. Next what, you're going to link to an article that says Sam Altman is also a homosexual, and we must be morally outraged at that next? Because that is where your direction is going, instead of arguing about the ill effects of AI.
I share the sentiment but the question here is how long can we maintain the balance point where the human in the agentic loop is required. It might be a window lasting only a few years, or for the foreseeable future; I think the answer lies the opaque compute economics of the frontier lab: how well models keep scaling and how economically sustainable is serving those models under the current market conditions.
If my job gets automated, I'll find something else to do. I wouldn't have wanted lamplighters to succeed in preventing electrification, so it would be unfair for me to prevent the automation of my job if it can be done.
I use Astra to drive Fable; I drivel into my phone while walking in the forest and it builds. I don’t need to check; I do as clients need to pay, but it always is great. And it surprises me with things I did not know were possible even (never encountered them before so why would I know). We are at the point where our clients send voice messages and they get what they want without humans basically. This costs 10+10 max2 subs but that’s nothing compared to hiring people. We didn’t fire anyone; we just have 100+ more clients and make almost 50x more money. It’s boring but great as long as it lasts, we are already where you say for what enterprises generally need for the boring parts. That’s 99%.
I find I still add value, but I don't know if my value is real. Am I just biased and expect things to be done a certain way and penalize the model for doing something different? and am I providing the model with enough high quality context to align with my expectations in one-shot?
These people need to take a vacation and come back in a few months when the vision models get better/cheaper.
Astra is already good at taking screenshots and acting on it (part of the agi claims).
What about cases where a bug is only obvious in motion? Or where you're working with a SBC and the bug is that you forgot a cable? Screenshots alone aren't enough. The authors point stands.