Anthropic says these topics are too dangerous to let its Fable 5 model talk about
- Thread starter JournalBot
- Start date
This pricing structure is quite absurd when you think about. Anthropic has control over the output tokens, and they charge more for those. They can fine tune exactly how verbose the model output is, and obviously the more verbose the more money they make. Opus 4.6 was very succinct. Opus 4.8 is extremely verbose. I tried 4.8 and immediately went back to 4.6 because it was burning through tokens a lot faster.
The "seller" is deciding how much product to sell you. Google search doesn't charge you by how many results your query returns.
Upvote
23
(23
/
0)
As usual, the billionaire class is keeping the sarin nerve gas and pipe bombs all for themselves.
Upvote
10
(11
/
-1)
I just tried this out on a research project that compares pathogen profiles of samples across various geographic regions, and it flagged an innocent stats question as a violation of their usage policy. It wasn't even referencing pathogen species, just broad categories (bacteria, fungi, etc.) Seems like this model is totally useless to me.
Upvote
14
(14
/
0)
Upvote
26
(26
/
0)
I don't know, honestly. I could see how this could be a real bummer, definitely. But this also feels to me like an advertising ploy. Like how B-movie theaters used to make you sign a form indemnifying them if ZaAt scared you to death...Or actually, maybe it's even more like how that rabbit doesn't want the kids to get its (sugary, unhealthy, tasteless) breakfast cereal...but then again, maybe I'd feel differently were I in cybersecurity.
Upvote
7
(7
/
0)
Heh, Ars comments ... you are always there when I need to depart the hype train and remember the cost of the ticket.
Upvote
16
(16
/
0)
You addicted to seeing how the ultra rich build their bunkers too?
Upvote
1
(1
/
0)
The alternative is corporate or billionaire gatekeepers. Sigh… the government was all we had. I fear the future will be quite dark.
Upvote
4
(4
/
0)
Anthropic: We were in the nick of time. You were in great peril.
Coder: I don't think I was.
Anthropic Yes, you were. You were in terrible peril.
Coder: Look, let me go back in there and face the peril.
Anthropic: No, it's too perilous.
Coder: Look, it's my duty as aknight security coder to sample as much peril as I can.
Anthropic: No, we've got to find the Holy Grail (of AGI). Come on.
Coder: Oh, let me have just a little bit of peril?
Anthropic: No. It's unhealthy.
Coder: I don't think I was.
Anthropic Yes, you were. You were in terrible peril.
Coder: Look, let me go back in there and face the peril.
Anthropic: No, it's too perilous.
Coder: Look, it's my duty as a
Anthropic: No, we've got to find the Holy Grail (of AGI). Come on.
Coder: Oh, let me have just a little bit of peril?
Anthropic: No. It's unhealthy.
Upvote
26
(26
/
0)
How many tiomes will you folks and other tech media writers just take all these prerss reelases as fact.
Good lord.
This product is reaaaaaaaly dangerous guyz, but we'll let a few trusted folks have it!.
Good lord.
This product is reaaaaaaaly dangerous guyz, but we'll let a few trusted folks have it!.
Upvote
16
(16
/
0)
The way I see it is that model advances are getting more and more expensive. Recent advances have increasingly relied on scaling up rather than core architectural improvements. Mythos has more than 10 trillion parameters. Compare this to GPT5's estimated 1.8 trillion parameters, or GPT-3.5's estimated 200 billion parameters. It's all about scale now.
The companies creating these models can't stop scaling up or they'll fall behind, but it's becoming tremendously expensive and unsustainable. Advances in compute hardware aren't happening anywhere near fast enough to support this explosion in scale. The only way to stop the feedback loop is for someone to intervene and force them to slow down. And the best possible case for that happening is on "safety grounds".
It almost seems they want it both ways – their whole shtick is about making ever more capable models, but at some point, they need regulators to step in and save the whole industry from imploding in a self-destructive cycle of unsustainable model scaling. If that can happen on "safety grounds" because the models are "too powerful", then that's a better story than simply admitting that scaling has run its course.
Upvote
13
(13
/
0)
I haven’t done chemistry professionally in 25 years, but it seeming like the field may have some life in it yet!
Upvote
1
(1
/
0)
So, basically, some dumbass snuck a copy of The Anarchist’s Cookbook into Mythos’s training set, it’s far too expensive to take it out now, and Anthropic PR is spinning this is as corporate responsibility?
Upvote
20
(20
/
0)
Markets whipsaw as AI sell-off resumes
https://edition.cnn.com/2026/06/09/investing/nasdaq-sp500-dow-drop-aiI must be the proverbial bellboy because I was just thinking it's time to buy in.
Upvote
4
(4
/
0)
Editing quotes is problematic, at best.
I fully understand the humor, point, and intent behind your discourse but other people might report it because it's dead seriously not appropriate.
If I have to clarify the previous post, ''either too dangerous for public use or absolutely fucking useless" is absolutely still accurate in economic and social terms. There's definitely use cases for it in the academic and backend infrastructure spaces, but just letting the general public suck down tokens or flagrantly display it's had more bad faith use cases than good ones since 2018 is still proving to have been the dumbest Pandora's Box ever, and the proof is in the pudding on both the token expense and public opinion polling.
You can pick nits over absolutist wording or insist on vague sarcasm about middle ground cases all you want, but they're going to drive the next Great Depression from sucking up capex from everything else for a shitty 8-ball that tries to murder people when you reject it's Git requests.
Upvote
4
(4
/
0)
Upvote
1
(1
/
0)
No, just hold on a bit longer, so we get better models that we can run locally! (well, -ish). THEN everything can burn down.
Upvote
0
(0
/
0)
Creationists, and their children, need to be protected from the science of evolution lest they lose their fierce convictions by accidentally discovering reality. /S
Upvote
-1
(3
/
-4)
But our latest model is vastly superior to our previous vastly superior model! It even passes tests that we created specifically for allowing it to pass! Progress! Our profits are up nearly 1,600%.*
*Matching the Orange Shitgibbon's "vastly superior" medical prescription savings, which the most moronic and innumerate GQP republicans still try to defend.
Upvote
8
(9
/
-1)
Irony is an alien concept to obscurantists. One very large company that I worked for (near 100K employees) had an IT department that totally ignored a worm that was ravaging the internet at the time. After 3 days of total computer systems shutdowns from the worm, IT finally recovered. And then they proceeded to give themselves great awards for what they had done. I don't think that the clueless C-suit management ever caught on. After all, it is only fair that the perpetrators get to award themselves for a job well done.
Upvote
6
(6
/
0)
It feels like we're stuck in a news cycle loop.
10: Anthropic claims (but provides no evidence) their model is soooo dangerous that mere plebs can't handle the responsibility that comes with such enormous power
20: America wins the totally-not-a-war with Iran... again. (Are you sick of winning yet?)
30: goto 10
10: Anthropic claims (but provides no evidence) their model is soooo dangerous that mere plebs can't handle the responsibility that comes with such enormous power
20: America wins the totally-not-a-war with Iran... again. (Are you sick of winning yet?)
30: goto 10
Upvote
12
(12
/
0)
Whether the Mythos actually has the capacity that they've claimed in their marketing questions is certainly dubious, but i'm not sure I agree with this take either.
The idea that companies uniformly "don't release products if they're worried about its safety" is pretty laughable. The US has multiple lawsuits progressing because BigTech ghouls literally did not care that their platforms elicited suicidal thoughts in children. Actually, worse - based on the evidence submitted execs saw that as a marketable insecurity, especially for teenage girls.
Upvote
9
(9
/
0)
So I've come to what's likely going to be fairly common conclusion about all this stuff:
When the AI companies are responsible for generating all content (not just on the internet but everywhere), they've will have also become the de facto censors for all content everywhere too. Awesome.
I had a run in with this myself recently. I decided to pay for a single month with the runway video generation service, just to give it a go and in this particular case, see if i could generate a video that comments on the elimination of trans-rights in the UK.
I put in a non-risque prompt (which was later confirmed to be non-risque by the support person I elevated a complaint to) and used their agent to help plan out the shots I wanted to create, all for it to block me when I tried to actually get it to generate the thing.
In my particular case, I was effectively using the service as a extremely cheap film crew, but tell me, how does a film crew that won't do what you want it to do have any use?
I've still got credits left to use up, but after that experience I haven't logged back on (I've been keeping the tab open to remind myself that I should use them up, but my enthusiasm is severely lacking).
I can't say I've got any interest in going back to the site. Why would i? It's a waste of time if it won't output what I want it to output because my creative vision gets filtered through their lens of what's acceptable.
Obviously that doesn't mean AI won't have it's uses, but I don't remember everyone on the planet getting together and agreeing that they could consume all our free and open data, just to turn around and charge us for a heavily censored product.
Ad-tagline: "Say hi to your new AI buddy!"*
* That will only do things it's want to do and if you don't like it, don't worry, we can subtly bias the information you receive from it over the next few months of usage, so eventually it won't even occur to you to want to do the things you originally wanted to do.
When the AI companies are responsible for generating all content (not just on the internet but everywhere), they've will have also become the de facto censors for all content everywhere too. Awesome.
I had a run in with this myself recently. I decided to pay for a single month with the runway video generation service, just to give it a go and in this particular case, see if i could generate a video that comments on the elimination of trans-rights in the UK.
I put in a non-risque prompt (which was later confirmed to be non-risque by the support person I elevated a complaint to) and used their agent to help plan out the shots I wanted to create, all for it to block me when I tried to actually get it to generate the thing.
In my particular case, I was effectively using the service as a extremely cheap film crew, but tell me, how does a film crew that won't do what you want it to do have any use?
I've still got credits left to use up, but after that experience I haven't logged back on (I've been keeping the tab open to remind myself that I should use them up, but my enthusiasm is severely lacking).
I can't say I've got any interest in going back to the site. Why would i? It's a waste of time if it won't output what I want it to output because my creative vision gets filtered through their lens of what's acceptable.
Obviously that doesn't mean AI won't have it's uses, but I don't remember everyone on the planet getting together and agreeing that they could consume all our free and open data, just to turn around and charge us for a heavily censored product.
Ad-tagline: "Say hi to your new AI buddy!"*
* That will only do things it's want to do and if you don't like it, don't worry, we can subtly bias the information you receive from it over the next few months of usage, so eventually it won't even occur to you to want to do the things you originally wanted to do.
Upvote
5
(5
/
0)
no
It is simply a way for the US government and US companies to use these softwares as a weapon against other countries, particularly European ones.
They decided to keep these developments to themselves, on the pretext of protecting against ‘bad actors’.
This serves as a reminder that we need to sever our ties with the United States and develop our own IT industries – something I have been hoping for for two decades now...
A powerful European IT industry with considerable resources would be beneficial for us (the loathsome foreigner who isn’t American…), and beneficial for IT (more choice, new hardware and software)
To be honest, we should be far more nationalistic and stand up to both the United States and Russia. These ‘AI’ systems prove ONCE AGAIN that the United States is working against us.
It is simply a way for the US government and US companies to use these softwares as a weapon against other countries, particularly European ones.
They decided to keep these developments to themselves, on the pretext of protecting against ‘bad actors’.
This serves as a reminder that we need to sever our ties with the United States and develop our own IT industries – something I have been hoping for for two decades now...
A powerful European IT industry with considerable resources would be beneficial for us (the loathsome foreigner who isn’t American…), and beneficial for IT (more choice, new hardware and software)
To be honest, we should be far more nationalistic and stand up to both the United States and Russia. These ‘AI’ systems prove ONCE AGAIN that the United States is working against us.
Last edited:
Upvote
2
(4
/
-2)
The whole guardrail argument is counterproductive to the use of AI. I rely heavily on claude for our company code reviews and rapid prototyping software solutions. Up till 4.6 we didn’t have any problems . 4.8 started to show spontaneous API errors for Access violations and AUP violations when working on our own code review. It became unusable.
Tried Fable 5 today, asked it to do a basic MR review for known npm package infections. Worked 5 min , spent $15 in api tokens , and told me it has API error for security violations.
Only way around it is to exit session, Then downgrade to 4.6 to complete the work.
It’s completely broken all workflows . Next management meeting this week we will be considering competitive alternatives to Anthropic.
Tried Fable 5 today, asked it to do a basic MR review for known npm package infections. Worked 5 min , spent $15 in api tokens , and told me it has API error for security violations.
Only way around it is to exit session, Then downgrade to 4.6 to complete the work.
It’s completely broken all workflows . Next management meeting this week we will be considering competitive alternatives to Anthropic.
Upvote
-1
(1
/
-2)
Upvote
1
(1
/
0)
"The company says it will be periodically expanding its existing Project Glasswing program “in consultation with the US government”"
They now face the same problem as European NATO generals...
They now face the same problem as European NATO generals...
Upvote
0
(0
/
0)
Wildcard files for the war-region and AI company, and you're good to go full Nostradamus...
Upvote
0
(0
/
0)
I genuinely can't use Fable for work. I'm a medical physicist. I use the word nuclear a lot (although usually in the context of NMR / MRI and occasionally in reference to relativistic particles). Opus is fine (well, 99% of the time - I've certainly hit the CBRN filters a few times and even been invited to email anthropic about the false positives).
Fable has literally refused to work on any of my problems (even those about fluid dynamics!) and just tells me that I'm violating anthropic's AUP. I've reached out to their support and don't expect to hear anything sensible back. One thing I do look forward to though is OpenAI offering an equivalent model but with less safeguards...
Fable has literally refused to work on any of my problems (even those about fluid dynamics!) and just tells me that I'm violating anthropic's AUP. I've reached out to their support and don't expect to hear anything sensible back. One thing I do look forward to though is OpenAI offering an equivalent model but with less safeguards...
Upvote
3
(3
/
0)
Biology? Chemistry? Damn, I must be a menace to society everyday and not realize it
Upvote
2
(2
/
0)
Always has been
If anything, I'd thought they were worried about the limits of scaling and getting more creative lately, but evidently that's not the case
Upvote
2
(2
/
0)
(You should not post when drunk, even with factual info.)
It is getting so super-intelligent that we can't let you find out just how intelligent it is! Only our lowly vibe coders can ever measure that because the knowledge could destroy anyone skilled in detecting so much
*An accidentally factual claim, the bar is that low.
Upvote
5
(5
/
0)
More to the point, China, presumably Russia, and others (Israel?) will create LLM’s with nefarious? capabilities that exceed Anthropic and OpenAI and are more readily available. Therefore is this another case of ‘race to the bottom?’
Upvote
2
(2
/
0)
They're doing it again. "Our product is so advance it'll break the internet."
It's hype. It was hype last time when they said their prototype hacked the whole internet, and it's hype this time.
It's hype. It was hype last time when they said their prototype hacked the whole internet, and it's hype this time.
Upvote
8
(8
/
0)
MilanKraft
Ars Tribunus Angusticlavius
Jeeezus. Thanks for posting this.
I knew they were stretching the truth at the time, and in the days immediately after people (including respectable outlets like Ars and the Economist) were like, "well, we spoke to some third party experts and seems like this might actually be a thing".... and this is the problem with the information environment social media companies and others have created for us.... basically there's no time anymore to just let things sit for a bit so people can gather info and draw conclusions.
This full-on lie you're talking about (I'm sure other aspects of their claim were similarly inflated) likely was discovered a week or more after the Mythos hype had started. And by then ~ 80% of the reading public had moved on (as we always do), accepting all Anthropic's claims as truth and "wow these new apps are super-powerful-amaze-balls!"
Pretty fucking sad when the ostensibly "ethically more careful" LLM company, is basically manufacturing huge security / "our model so powerful, bro!!¡¡" claims out of literal thin air. Even more sad when journalists don't have the time to find out properly "is this even sorta true" before reporting it to the world. Because if they try to do that, by the time they're ready to report the world has literally moved on.
This is a very dangerous place for society to be right now, and hugely magnified by all the populist, wannabe tyrant assholes taking positions of power in the world (especially here).
All of this is...... not great.
Misanthropic is the way.
Upvote
9
(9
/
0)
