I'm surprised how much of the discussion here is taking the angle of "Anthropic pretended this model was soooo dangerous for months as marketing, and it seems like someone decided to believe them!"
First, Anthropic was founded by people who we know were worried about AI safety and signs point to that still being the case. It's really cynical to say it was all an exaggeration for marketing.
Second, this isn't Moller promising a fantastic working flying car next year. The model did what Anthropic said it could do.
I realise that ruling out "they bought Anthropic's scaremongering" brings up the question of why the government would block Mythos/Fable, but not the roughly-as-capable and less restricted GPT5.5. However we do know for a fact that they dislike Anthropic more than OpenAI right now.
There are a lot of dangerous things in the world and surprisingly a lot of people can avoid the constant stream of chicken little nonsense.
If everyone expended the same amount of marketing effort trying to scare the ** out of everyone that Anthropic does, it'd be a very miserable world to live in.
We are unfortunately a captured audience and the autistic people at Anthropic are abusing this.
There have been a steady stream of articles about exactly this over the past few weeks.
Yesterday there was one about 5 zero-days in ffmpeg. Another commenter mentioned the fixes done to Firefox.
If you put a minor effort into looking for news about Mythos making security patches and fixing critical bugs in important projects recently, you will find them.
I don't see any reason we should put weight behind their supposed fears today though. It's completely irrational to build the very thing you think could seriously harm or kill us all.
Yes they may have had those fears before, but even then it didn't stop them from building companies and running full speed towards the end goal with little to no effort spent on meaningful safety efforts.
The idea that the government introduced export controls on it because they "fell for the marketing" is stupid. It's much more likely they're being vindictive. There's plenty of evidence that that's how the current government acts.
Little bit of column a, little bit of column b. A lot of the actions of the current USG seem to occur at the intersection of shared interests but different motivations, consider that:
- apocalypse-cult evangelicals (Mike Johnson types)
- secular RE development globalists (Kushner types)
- white supremacist / eugenicist weirdos (Stephen Miller types)
- SV / tech billionaire stooges (Vance types)
- media / propaganda old guard (Koch, Murdoch, Heritage)
- Morally bankrupt grifters / influencers (too many to mention)
all seem to somehow be under the same tent right now. Luckily for us, history points towards such unlikely alliances as being fragile and short-lived. Unluckily, when such alliances have gained power they usually don’t let it go without making sure lots of people suffer first.
Edit: I call it an unlikely alliance because there are represented many reactionary accelerationists who all have a different vision of what America should look like after the revolution.
> If they were money hungry they wouldn't have fought the DOW
I think it could be reputation management exercises. Especially how it was aligned with airstrike on Iranian girls elementary school and statements that Claude were picking targets.
> In the future they will come with their 'ready' solution, already 'working' and be even less receptive
This has already been common in the audio engineering world for some time, as home demos of music approach professional quality. As you forsee here, people get used to what they have and become even less inclined to accept changes in a new professional mix.
There was this video which showed film directors doing something similar. They film their scene with “filler” music previously used in some other film. Then they hand the scene to the music director and ask a score for it, which forces the director to make something very similar to the filler music. It then makes all film music sound similar.
This was only possible due to the “productivity boost” of digital editing pipelines, which allowed directors to edit immediately after filming.
Most people want to spend time with others in the same kind of way they want to eat food or sleep or watch a movie. It just seems to be built in. People who appear to have ulterior motives are treated suspiciously. Some people seem to need a lot more social time than others, but most people desire at least a little bit of human contact.
I used to have saving turned on for my MSN Messenger chats back around 2001-2004. I didn't lose them. 10-15 years later or so I had a look through them and the cringe was so powerful that I deleted them all anyway.
I always had enough storage to more or less never be concerned about deleting stuff. I have almost every backup of every phone, computer, relevant app, forum private messages, emails, and so on going back ~25+ years. I found my Yahoo Messenger chat logs and I had the exact same cringe reaction. I didn't delete them but one thing's for sure, my writing style and thinking have changed so much that those artifacts are unrecognizable.
This was very unexpected to me because in my mind the changes only happened as I became a young adult. The evidence of these decades of logs shows the change continued to happen as an adult.
I have my original gmail address from 20 years ago still and even old youtube videos my friends and I uploaded from ~18 years ago.
The cringe is rough but at some point the cringe becomes so bad it loops back around to me just feeling nostalgic and grateful that there's proof I was able to do things, create, be silly, whatever without worrying about appearances so much.
Also, I figure if I ever become a megalomaniac then old youtube videos of my teenage self doing parkour should go pretty far in humbling me (although, honestly, I think 13 year old me was way cooler than I am now, so I guess it could backfire).
I'm in a similar boat, I previously saved a bunch of AIM chats from my school days but only tiny fragments made it to present day. From those fragments I gather I wouldn't want to re-read much of it. :D
In Yuval Noah Harari’s brilliant new book 21 Lessons for the 21st Century, he writes that Artificial Intelligence, with its limitless potential and connectedness, will ultimately render many humans redundant in the work place. This sounds entirely feasible. However, he goes on to say that AI will be able to write better songs than humans can. He says, and excuse my simplistic summation, that we listen to songs to make us feel certain things and that in the future AI will simply be able to map the individual mind and create songs tailored exclusively to our own particular mental algorithms, that can make us feel, with far more intensity and precision, whatever it is we want to feel. If we are feeling sad and want to feel happy we simply listen to our bespoke AI happy song and the job will be done.
But, I am not sure that this is all songs do. Of course, we go to songs to make us feel something – happy, sad, sexy, homesick, excited or whatever – but this is not all a song does. What a great song makes us feel is a sense of awe. There is a reason for this. A sense of awe is almost exclusively predicated on our limitations as human beings. It is entirely to do with our audacity as humans to reach beyond our potential.
It is perfectly conceivable that AI could produce a song as good as Nirvana’s Smells Like Teen Spirit, for example, and that it ticked all the boxes required to make us feel what a song like that should make us feel – in this case, excited and rebellious, let’s say. It is also feasible that AI could produce a song that makes us feel these same feelings, but more intensely than any human songwriter could do.
But, I don’t feel that when we listen to Smells Like Teen Spirit it is only the song that we are listening to. It feels to me, that what we are actually listening to is a withdrawn and alienated young man’s journey out of the small American town of Aberdeen – a young man who by any measure was a walking bundle of dysfunction and human limitation – a young man who had the temerity to howl his particular pain into a microphone and in doing so, by way of the heavens, reach into the hearts of a generation. We are also listening to Iggy Pop walk across his audience’s hands and smear himself in peanut butter whilst singing 1970. We are listening to Beethoven compose the Ninth Symphony while almost totally deaf. We are listening to Prince, that tiny cluster of purple atoms, singing in the pouring rain at the Super Bowl and blowing everyone’s minds. We are listening to Nina Simone stuff all her rage and disappointment into the most tender of love songs. We are listening to Paganini continue to play his Stradivarius as the strings snapped. We are listening to Jimi Hendrix kneel and set fire to his own instrument.
What we are actually listening to is human limitation and the audacity to transcend it. Artificial Intelligence, for all its unlimited potential, simply doesn’t have this capacity. How could it? And this is the essence of transcendence. If we have limitless potential then what is there to transcend? And therefore what is the purpose of the imagination at all. Music has the ability to touch the celestial sphere with the tips of its fingers and the awe and wonder we feel is in the desperate temerity of the reach, not just the outcome. Where is the transcendent splendour in unlimited potential? So to answer your question, Peter, AI would have the capacity to write a good song, but not a great one. It lacks the nerve.
IMO Morse code misses the point, because it doesn’t depict letters, it encodes the abstract characters into completely different “glyphs”. In particular, knowing the English letters doesn’t make you understand Morse code the way it makes you able to read Two Slice; you need to know the mapping between letters and Morse code.
You can trivially encode any alphabet into a one-dimensional graphical encoding. But you couldn’t, for example, have a remotely intelligible two-pixel Chinese font in the way Two Slice demonstrates that you can have an English one.
I have really noticed recently that a lot of modern media (film, TV, videogames, etc) seems much more based on prior media than on the author's experience of the world. Like everything is now operating at a meta level. It's a little sad.
I wrote a response to this, but then I realised I was responding to the claim that modern media was more derivative, rather than what you actually said, which was that modern media is more _meta_.
Can you go into that a little more? Do you have specific examples that make you sad?
The first example that comes to my mind is the show Community, which I really enjoy, and which doesn't make me sad at all.
P.S. an article I linked to in my original response was https://www.filfre.net/2025/01/the-crpg-renaissance-part-1-f... which I mentioned as it talks about a historical standout in the genre but puts it in the context of the copycats and the schlock. It's now irrelevant to my comment, but I'd like to link to it anyway.
Not OP, but there is a wide chasm between what Community does and what OP was referring to.
Community's thing is that it is a meta show. It uses the meta it references to get a point across, make a joke, or provide a spectacle (a good example of spectacle are the Paintball episodes)
What OP referred to, and what I've noticed, was that media nowadays is just a mashup of what came before with little to say about it. Or to put it in other words: not transformative. The creator likes something, and they put it in their work because it's cool. There's nothing wrong with doing just that, but when you start seeing the same thing over and over again in different works, it gets tiresome.
We're so obsessed with filling every waking moment with something that we don't allow ourselves to have the "a-ha!" moment any more, so we default to "what if X and Y?" where X and Y are thoughts on the surface of our mind rather than two unrelated things that somehow click when the default mode network activates. For example: what do archways in a Shinto shrine have to do with a fox piloting a starship around? Absolutely nothing, and yet for Miyamoto that thought made sense.
Ah, thank you very much for this reply, because I haven't watched Community myself so I didn't realise the confusion between a show that's intentionally about a meta situation vs. ... well what you've written explains my meaning exactly.
You should watch some episodes, even if you don't watch all of it. There's a reason why it influenced popular culture (even if no one remembers it doing so).
I don't know if I have a good argument for it myself. I have seen a lot of people saying specifically that they based their {thing} on {prior thing} rather than something from life, but I haven't exactly kept a list. Beyond that it's mostly a feeling.
The very brief (and bastardised) summary is that we're cutting ourselves from what is real, so we base our art on the fake reality that we're experiencing.
I'll never forget when one of my teachers asked: "who has seen a sheep?" The entire class put up their hand. The next question was "who has seen a live sheep, in front of them?" more than half the class put their hand down. We all know what a sheep looks like, but not because we've been near one.
Yes indeed, I'm aware of it, though I admit I never finished the whole thing. It did make me notice this situation even more acutely.
It's funny that the part everyone quotes from the book (namely the Borges fable and the 'desert of the real itself') is in the introduction. Makes me wonder how many others didn't actually get through it. :)
With the level of ability that AI is at right now, I've found it useful personally to think of it something like a very good search over existing knowledge. Another step up in searchability in the lineage of reference books, stack overflow, GitHub etc.
Programmers are rewriting and reinventing the same techniques more often than any other vocation I can think of, and so we were primed for a really good search over prior art. The fact that AI can also adapt that prior art to your particular use case makes it even more powerful.
Much like how great success never came from cobbling together various bits of copy-pasted code from Stack Overflow though, current AI can't really build your whole project.
Nice reusable libraries are still a core part of most AI projects, but honestly I think it's not a terrible approach with all the updating dependency malware issues with stuff like NPM.
> I think it's not a terrible approach with all the updating dependency malware issues with stuff like NPM
I think in this instance, the only thing worse than a zero day in your dependency tree, is a zero day you don't know your LLM vendored directly into your codebase...
Personally I feel a vulnerability in local code (unshared ai slop) is much less likely to be exploited, than for say a npm package update that will pwn you as soon as it loads up.
And on other hand I really do not understand how basic project boilerplate templating wasn't already a fully solved issue. Surely it should have been doable...
I guess the nice thing about AI is it points to the things that we really need to figure out how to abstract instead of rewriting from scratch all the time.
I sort of hold the opposite view: pretty much anything I want to do, there are about ten competing libraries/frameworks/languages, and the lack of commonality across them means I'm often wedged into weird ecosystem choices
Yes, I don't have anything important to say other than I 100% agree with this comment. AI in its current state is akin to Stack Overflow and Google on steroids, but from my experience, it doesn't do well building out full-scale applications other than perhaps some initial scaffolding.
If I were to use it against a legacy, rather poorly written codebase, where the code may be hard to understand without some in-depth analysis. I could certainly ask an AI agent to read the code (How does application X do Y, for example), but I wouldn't have it start hammering out features or have it do any type of refactoring. That would cause far too many commits and confusion amongst the development team, leading to even more slop than whatever we'd already be dealing with.
Just leaving this comment here so I can come back to your comment. I've been getting a bit discouraged by AI lately, but this sums up my experience with it well enough.
> Yes, I don't have anything important to say other than I 100% agree with this comment. AI in its current state is akin to Stack Overflow and Google on steroids, but from my experience, it doesn't do well building out full-scale applications other than perhaps some initial scaffolding.
We're currently using it to build out a full-scale application. It does as well as you care to coax into doing tbh. You have to invest heavily in harness engineering, and at least my experience has been that as you do that, the results improve.
>It does as well as you care to coax into doing tbh. You have to invest heavily in harness engineering, and at least my experience has been that as you do that, the results improve.
That is also my experience.
When starting a project I observe how the agent fails, I add new rules to the harness to prevent it from falling and repeat the process until I am happy with the output.
I can't point you to a good complete documentation, because the field is changing very fast as people make new discoveries.
I learned by reading articles, success stories failure stories and mostly by doing, trying stuff, see how it works and adjusting it and burning a lot of tokens along the way.
What I would do in your shoes, I would ask an AI chat to find new articles on the matter (including on HN), explain how Codex, Claude, Pi are managing agents.
My compressed view is: you need to have a great specification both business and architecture wise that doesn't leave anything important for the model to guess because chances are it will make the wrong choices. That comprehensive spec should not be in one huge chunk. Have your plan divided in phases that each fit in a context window and have the spec for each phase.
Use TDD, strive for 100% coverage.
Force the model to behave: if it doesn't do what is supposed to, give it feedback and ask it to retry and don't allow it to progress to the next stage unless everything is perfect.
I also like to write comprehensive integration tests before building anything. The agents are not allowed to touch or read the integration tests, only run them and they will get feedback where the tests fail. I like to build the integration tests in a different language than the software I am building, to make sure there isn't something platform specific that the tests rely on. I use C#, Go, Rust and Zig for development and Python for the integration tests.
For now, to get good results, I can't just copy and paste the setup from a project to another, I have to work a lot to tailor the process for each new codebase.
And that's why I am working on an agent harness to try to force the agents to do the right things in most common development scenarios without wasting much tokens. By common development scenarios I mean that is a large goal, right now I am working towards backend web development and microservices.
In my experience, you’ll eventually hit a context window issue and it will just start spouting gibberish/doing wrong things, and nothing will significantly improve it. But hey, maybe it’s improved.
Well, auto-compaction is a thing in Claude Code now. Plus we have /goal command and some automated review stuff, so you can kinda just get it to loop until the automated reviews are satisfied and CI is passing. Does most of the heavy lifting.
reply