The LLM generated the response that was expected of it. (statistically)
And that's a function of the data used to train it, and the feedback provided during training.
It doesn't actually have anything at all to do with
---
"It generated a take-down style blog post because that style is the most common when looking at blog posts criticizing someone."
---
Other than that this data may have been over-prevalent during its training, and it was rewarded for matching that style of output during training.
To swing around to my point... I'd argue that anthropomorphizing agents is actually the correct view to take. People just need to understand that they behave like they've been trained to behave (side note: just like most people...), and this is why clarity around training data is SO important.
In the same way that we attribute certain feelings and emotions to people with particular backgrounds (ex - resumes and cvs, all the way down to city/country/language people grew up with). Those backgrounds are often used as quick and dirty heuristics on what a person was likely trained to do. Peer pressure & societal norms aren't a joke, and serve a very similar mechanism.
> Like most things - assume the "20/100/200" dollar deals that are great now are going to go down the enshitification route very rapidly.
I don’t assume this at all. In fact, the opposite has been happening in my experience: I try multiple providers at the same time and the $20/month plans have only been getting better with the model improvements and changes. The current ChatGPT $20/month plan goes a very long way even when I set it to “Extra High” whereas just 6 months ago I felt like the $20/month plans from major providers were an exercise in bouncing off rate limits for anything non-trivial.
Inference costs are only going to go down from here and models will only improve. I’ve been reading these warnings about the coming demise of AI plans for 1-2 years now, but the opposite keeps happening.
> Inference costs are only going to go down from here and models will only improve. I’ve been reading these warnings about the coming demise of AI plans for 1-2 years now, but the opposite keeps happening.
This time also crosses over with the frontier labs raising ever larger and larger rounds. If Anthropic IPO (which I honestly doubt), then we may get a better sense of actual prices in the market, as it's unlikely the markets will continue letting them spend more and more money each year without a return.
> The current ChatGPT $20/month plan goes a very long way
It sure does and Codex is great, but do you think they'll maintain the current prices after/if it eventually dominates Claude Code in terms of marketshare and mindshare?
At this point subsidizing Chinese open-weights vendors by paying for them is just the right thing to do. Maybe they too might go closed-weights when they become SotA, but they're now pretty close and haven't done it.
The harness is effectively the agent's 'body'. Swapping the brain (model) is good, but if the body (tools/environment) is locked down or inefficient, the brain can't compensate. Local execution environments that standardize the tool interface are going to be critical for avoiding that lock-in.
I have a lot of respect for Canonical for driving a distro that was very "noob friendly" in an ecosystem where that's genuinely hard.
But I mostly agree with you. Once you get out of that phase, I don't really see much value in Ubuntu. I'd pick pretty much anything else for everything I do these days. Debian/Fedora/Alpine on the server. Arch on the desktop.
Yeah, the folks in here recommending Debian as a solution to this problem are insane.
I love Debian, it's a great distro. It's NOT the distro I'd pick to drive things like my laptop or personal development machine. At least not if you have even a passing interest in:
- Using team communication apps (slack/teams/discord)
- Using software built for windows (Wine/Proton)
- Gaming (of any form)
- Wayland support (or any other large project delivering new features relatively quickly)
- Hardware support (modern linux kernels)
I'd recommend it immediately as a replacement for Ubuntu as a server, but I won't run it for daily drivers.
Again - Arch (or it's derivatives) are basically the best you can get in that space.
I think Debian Stable, Ubuntu LTS, and derivatives thereof are particularly poor fits for general consumers who are more likely to try to run the OS on a random machine they picked up from Best Buy that’s probably built with hardware that kernels any older than what ships in Fedora are unlikely to support.
The stable/testing/etc distinction doesn't really help, either, because it's an alien concept to those outside of technical spheres.
I strongly believe that the Fedora model is the best fit for the broadest spread of users. Arch is nice for those capable of keeping it wrangled but that's a much smaller group of people.
I'll add - I think the complexity is somewhat "over-stated" for Arch at this point. There was absolutely a period where just reading the entire install guide (much less actually completing it) was enough to turn a large number of even fairly technical people off the distro. Archinstall removed a lot of that headache.
And once it's up, it's generally just fine. I moved both my spouse and my children to Arch instead of Windows 11, and they don't seem particularly bothered. They install most of their own software using flatpaks through the store GUI in Gnome, or through Steam, the browser does most of the heavy lifting these days anyways.
I basically just grab their machine and run `pacman -Syu` on it once in a while, and help install something more complicated once in a blue moon.
Still requires someone who doesn't mind dropping into a terminal, but it's definitely not what I'd consider "all that challenging".
YMMV, but the issue I usually run into with Arch is that unless you watch patch notes like a hawk, updates will break random things every so often, which I found quite frustrating. The risk of this increases the longer the system goes without updates due to accumlated missing config file migrations and such.
Even as someone who uses the terminal daily it's more involved than I really care for.
I agree that they are a poor fit for a random user especially for debian install being not as intuitive but for supporting hardware I disagree.
I decided to try debian stable on my brand new gaming PC and it worked fine out of the box. Combine with steam flatpak for gaming and I have less issues than my friends who game on Arch.
I agree though that Fedora is probably a good general recommendation.
Over time I evolved to Debian testing for the base system and nix for getting precise versions of tools, which worked fairly well. But, I just converted my last Debian box to nixos
I'm using Debian testing in my daily driving desktop(s) for the last, checks notes, 20 years now?
Servers and headless boxes use stable and all machines are updated regularly. Most importantly, stable to stable (i.e. 12 to 13) upgrades takes around 5 minutes incl. final reboot.
I reinstalled Debian once. I had to migrate my system to 64 bit, and there was no clear way to move from 32 to 64 bit at that time. Well, once in 20 years is not bad, if you ask me.
I've had a couple outages due to major version upgrades: the worst was the major version update that introduced systemd, but I don't think I've ever irreparably lost a box. The main reason I like nixos now is:
1) nix means I have to install a lot fewer packages globally, which prevents accidentally using the wrong version of a package in a project.
2) I like having a version controlled record of what my systems look like (and I actually like the nix language)
I prefer to isolate my development environment already in various ways (virtualenv, containers or VM depending on the project) so I don't need that parts of NixOS. My systems are already run on a well-curated set of software. Two decades allowed me to fine tune that aspect pretty well.
While I understand the gravitas of NixOS, that modus operandi just is not for me. I'm happy and fine with my traditional way.
However, as I said, I understand and respect who use NixOS. I just don't share the same perspective and ideas. Hope it never breaks on you.
Currently Debian wants to deprecate GTK2. So even the guys that are interested in stability might start to see problems with Debian. The key problem of Linux is that it doesn't have a stable API to write long living GUI-software for. So far Debian was the way to go. Maybe recommending Debian will become even less popular soon.
You're allowed to throw debian testing or arch in a chroot. The only thing that doesn't work well for is gaming since it's possible for the mesa version to diverge too far.
Debian has multiple editions, if you want Arch, go for sid/testing.
Stable is stable as in "must not be broken at all costs" kind of stable.
basically everything works just fine. there's occasionally a rare crash or gnome reset where you need to login again, but other than that not many problems.
There are times where there are known bugs in Debian which are purposely not fixed but instead documented and worked around. That’s part of the stability promise. The behaviour shall not change which sometimes includes “bug as a feature”
Again, I like Debian a lot as a distro (much more than Ubuntu), but it's just not the same as a distro like Arch, even when you're on testing. Sid is close, but between Arch and sid... I've actually found fewer issues on Arch, and since there's an existing expectation that the community maintains and documents much of the software in AUR, there's almost always someone actually paying attention and updating things, rather than only getting around to it later.
It's not that Debian is a bad release, but it's the difference in a game on steam being completely unavailable for a few hours (Arch) or 10 days (Debian testing) due to an upstream issue.
I swapped a while back, mostly because I kept hitting issues that are accurately described and resolved by steps coming from Arch's community, even on distros like Debian and Fedora.
---
The power in debian is still that Ubuntu has made it very popular for folks doing commercial/closed source releases to provide a .deb by default. Won't always work... but at least they're targeting your distro (or almost always, ubuntu, but usually close enough).
Same for Fedora with the Redhat enterprise connections.
But I've generally found that the community in Arch is doing a better job at actually dogfooding, testing, and fixing the commercial software than most of the companies that release it... which is sad, but reality.
Arch has plenty of its own issues, but "Stale software" isn't the one to challenge it on. Much better giving it a pass due to arch/platform support limitations, security or stability needs, etc... All those are entirely valid critiques, and reasonable drivers for sticking to something like Debian.
Not joking, Arch. Pick Gnome/KDE/Sway as you please.
Arch is a wonderful daily driver distro for folks who can deal with even a small amount of configuration.
Excellent software availability through AUR, excellent update times (pretty much immediate).
The only downside is there's not a ton of direct commercial software packaged for it by default (ex - most companies they care give a .deb or a .rpm) but that's easily made up for by the rest of AUR.
While I definitely appreciate that this exists now (as another person who considered matrix and ended up passing due to deployment complexity) this is not what I think most folks would reasonably call a "trivial" docker compose setup.
It's a 16 service compose setup, complete with init hacks, inline docker-file builds to use those init hacks, a whole bunch of required config templates, some services that aren't clear if they're examples or requirements (ex - why is mailhog in there? just give me the SMTP env vars), and just a lot of general complexity still.
This feels like several discrete services that don't play nicely, herded together like cats. It doesn't feel like a solid and planned set of tools.
---
From my end - it's not enough to just stand it up. If this is my primary messaging tool and I'm hosting it, I need to have a feel for how it might break, and how I can fix it when it does.
Hell, I'm not even allergic to k8s (I host dozens of services on a baremetal cluster), but I am allergic to helm for very similar concerns: Complexity at the self-hosting scale (individual to small business) is rarely worth the additional overhead, and helm rapidly makes what should be simple yaml file deployments a complex, abstracted process. Your docker compose has a similar feel.
My first rule of thumb is "How long will it take me to manually read and understand a compose file while converting it to a k8s deployment?" This one looks onerous, not trivial.
To the framing issue - I can frame an alternate lens through which we balance enrichment against engagement.
Media can enrich people - expose them to new ideas, new stories, different views and opinions. This expands worldview and generally trends in the same direction as education.
Media can also be engaging - Use tools that make it compelling to continue viewing, even when other things might be preferable, on the low end: cliffhangers and suspenseful stories. on the high end: repetitive gambling like tendencies.
I'd argue if we view tiktok through this lens - banning it seems to make sense. Honestly, most short form social media should be highly reviewed for being low value content that is intentionally made addictive.
---
It's not society's job to cater to the whims of fucking for-profit, abusive, media companies. It's society's job to enrich and improve the lives of their members. Get the fuck outta here with the lame duck argument that I need to give a shit about some company's unethical profit motives.
I also don't care if meth dealers go bankrupt - who knew!
I fundamentally don't think governments should do a careful cost-benefit analysis of everything in society and then ban it if it falls on the wrong side. Just on basic principles of personal freedom. That's why the "addiction" framing is so important, because it implies that citizens don't have agency, and so justifies the authoritarian intervention.
PS if we apply your analysis to video games they surely would have been banned too.
Edit: by the way I remember back in the day we searched for "addicting flash games" and it was seen as a positive ;p
It is completely unreasonable for a society to do a careful cost-benefit analysis of everything in society - it's completely reasonable for a society to identify highly harmful things (especially those that hijack our brains through direct chemical or emotional addiction) and police those, or, as per Portugal's approach, make available societal supports to allow people to better cope with that addiction. The later isn't very reasonable to expect in a world of rising austerity due to financialization so the former seems more realistic.
"Hijack our brains" - exactly what I mean by pretending people don't have agency. Who gets to decide what counts as hijacking and what is just normal culture? Anything is "hijacking" to some extent - boy bands hijack teen girl brains, the BBC created Teletubbies to hijack toddler brains, heck any artistic representation is a hijack to the extent that it is interpreted by your brain at least partially as something other than what it really is i.e. some colours on a flat surface. The point is a new form of culture, communication and coordination is emerging and the old powers are shitting their pants.
(Fully agree on the Portugal approach though. The difficult to accept answer is that if people are choosing a shit life of scrolling 10 hours a day maybe we should do the actual hard work of improving the kind of life open to them.)
I remember that website, it was called addictinggames.com and I remember finding that bad grammar offensive. (I was obviously a lot of fun at parties.)
With social media, the cost benefit analysis doesn't deliver marginal results, just less stark/concentrated results. Drink driving is self evidently bad even though 99 times out of 100(?) it does no harm, because one time out of a hundred its consequences are catastrophic. Social media on the other hand is harming essentially 100% of the population in initially milder ways - even if you don't use it you're forced to live in a dumbed down society where wealth and power is becoming concentrated in the hands of those who pedal digital dopamine and in a democracy being undermined by disinformation. Of course 'initially milder harm' is step one in frog boiling.
> * even if you don't use it you're forced to live in a dumbed down society where wealth and power is becoming concentrated in the hands of those [...] *
Exactly the same applies to TV but where is all the handwringing about that? Remember those stats about people watching 7 hours of TV a day? Those people need some serious help too. What's happening is clearly just the old mass-media-supported order refusing to yield power to new media used by younger people. Governments couldn't care one bit about false information[1], nor about zoomers getting brainrot, it's all about controlling the flow of information.
[1] ("disinformation", another nice example of framing which ignores the fact that people have agency)
edit: the system is escaping my asterisks automatically now, anyone know how to get italics now?
This somewhat falls apart the second you realize that current models are already choosing which tools to use all the time. You can argue that's not "desire" but I'm not sure you'd convince me.
Frankly - even the other end of your argument is weak. Humans don't particularly want to control air traffic either (otherwise why are we having to pay those air traffic controllers their salaries to be there?). They do it as a function of achieving a broader goal.
He spends a lot of words talking about how saving cognition is equivalent to saving resources, but seems to gloss over that saving money is also saving resources.
Given the token/$ exchange rates is likely only going to get better for actual money over time...
If his predictions come true it seems clear that if your software isn't free, it won't get used. Nothing introduces friction like having to open up a wallet and pay. It's somewhat telling that all of his examples of things that will survive don't cost money - although I don't think it's the argument he meant to be making given the "hope-ium" style argument he's pushing.
---
Arguably, this is good long term. I personally think SaaS style recurring subscriptions are out of control, and most times a bad deal. But I also think it leaves a spot where I'm not sure what sort of career will exist in this space.
I think "reputation" is absolutely critical to functional societies, and this feels a lot like putting a mask on and hiding critical information.
If Facebook got rejected because people hate Facebook, even when the economics are good... that's valuable to society as a feedback mechanism to force Facebook to be, well - not so hated.
Letting them put a legal mask on and continue business as usual just feels a bit like loading gunpowder into the keg - You make a conditions ripe for a much larger and forceful explosion because they ignored all the feedback.
---
Basically - the companies are fighting their reputations for good reason. People HATE them. In my opinion, somewhat reasonably. Why are we letting them off the hook instead of forcing them to the sidelines to open up space for less hated alternatives?
If I know "Mike" skimps on paying good contractors, or abuses his employees, or does shitty work... me choosing not to engage with Mike's business, even though the price is good, is a perfectly reasonable choice. Likely even a GOOD choice.
The argument was that people's collective judgment, given transparency, will result in good decisions.
But we see from the Nov 2024 elections (and others, but most glaringly that one), that that is, sadly, not true.
So the people rejecting Facebook because of Facebook's reputation tells you nothing about whether Facebook is bad, because the people could have just as easily been bad.
The problem is that many people liked what they saw. Reputation was still important, but there were different beliefs about what reputations were desirable.
The LLM generated the response that was expected of it. (statistically)
And that's a function of the data used to train it, and the feedback provided during training.
It doesn't actually have anything at all to do with
---
"It generated a take-down style blog post because that style is the most common when looking at blog posts criticizing someone."
---
Other than that this data may have been over-prevalent during its training, and it was rewarded for matching that style of output during training.
To swing around to my point... I'd argue that anthropomorphizing agents is actually the correct view to take. People just need to understand that they behave like they've been trained to behave (side note: just like most people...), and this is why clarity around training data is SO important.
In the same way that we attribute certain feelings and emotions to people with particular backgrounds (ex - resumes and cvs, all the way down to city/country/language people grew up with). Those backgrounds are often used as quick and dirty heuristics on what a person was likely trained to do. Peer pressure & societal norms aren't a joke, and serve a very similar mechanism.
reply