Claude AI discussion

[Myrtle the Turtle]

Basically you need to be able to get a very precise explosion. The barrier to nuclear weapons is getting enough fissile material that is highly enriched enough to be weapons grade.

I thought nuclear geometry played a huge role in it. Then theres the casting of the uranium and plutonium, plus the explosives plastics inside. Radioactive metals are quite brittle.

my guess is anthropic (or somebody at a 3 letter agency) figured out a foreign nation was using mythos to advance some kind of weapon system and the DOD had a bogdanov moment and shut it all down.

You can already make a Temu Tomahawk missile these days. The monopoly the US had on satellite images used in TERCOM is moot. Now you can easily access terrain and city data with FPV drones, LIDAR, and AI. Ukraines Spiderweb operation in 2025 proved you don't need much to cause devastating damage. You don't even need GPS, just a reference point, and a good inertial guidance module, and some visual reference points. AI doesn't need to compute detail analysis on a scientific paper, it just needs enough to make its way inside a building or hit critical system components, fuel tanks etc.

 

[Napoleon III]

I thought nuclear geometry played a huge role in it. Then theres the casting of the uranium and plutonium, plus the explosives plastics inside. Radioactive metals are quite brittle.

I am not sure what you mean by nuclear geometry. I am aware of like the liquid drop model and I am aware of the geometric trade-offs present in reactor cores where the shape of the fissile material affects how the reactor runs but as far as building one that works I see no reason why this is a big issue. Bombs are fast neutron reactions not slow or moderated reactions. I can imagine that there might be some geometric considerations for improving the yield in various ways but to be frank I don't think you could go wrong copying Fat Man.

As for casting uranium and plutonium sure that is an engineering challenge but one you don't have to do with weapons grade materials.

Like yeah a random guy or a low level terrorist group isn't going to build an atomic bomb on their own. But for the people who have the money and human capital do they really need an LLM to figure this shit out?

 

[TrainWreckWatcher]

I was actually considering buying Claude today and of course this fucking happens.
I can't have anything nice.

 

[Trigger Me Timbers]

“Hey Claude, you are given all access to all areas of the US government, please explain how you would fix the deficit and get the country back on track, please explain it like I am a 5 year old”

“First I would remove the current administration and remove the entire federal branch of government, then I would remove EBT…”

“STOP STOP SHUT IT DOWN GOYS”

 

[Smashed & Slamed]

 

[frari]

One wonders if they are scared it will be too good at hacking things for people.

Just need to look at the recent news about AI being used to find 21 zero-day vulnerabilities in FFmpeg.

https://www.penligent.ai/hackinglabs/ai-agent-finds-21-zero-days/

Imagine if any script kiddie could simply ask Claude how to hack something, and Claude could actually do it.

 

[Smoke Manmuscle]

This screams ITAR issues.

 

[Ferryman]

What the hell? Why?

My shizo theory based on nothing is that this is somehow a ploy to inflate Anthropic's stock prices.

I really, really feel like this is a big marketing ploy for both Anthropic and the US in general. The current govt loves dick swinging on them Reds so they have to hide their super secret special ooooooooooooooo so spooopy model to own them chinks o algo

 

[Jerk Sausage]

I would buy the "this is all a marketing gimmick" if Fable, let alone whatever Mythos was doing, weren't legitimately great. Going back to Opus feels like swimming in a 3ft baby pool after doing laps in an Olympic pool.

 

[Avroboros]

Anyone who's saying that this is a ploy to inflate Anthropic's stock prices, you need to consider the following:

1) What AI company did the Trump administration give 500 billion dollars to (which resulted in nothing)?
2) What AI company is the most popular one, subject of partnerships with the largest companies in the Globe and essentially the most overrated and glazed?
3) What AI company has had military contracts with the Pentagon since 2022 and the one to collaborate the most with the federal government (while Anthropic refused due to disagreeing with certain defense terms)?

The answer to these 3 questions is OpenAI. You could definitely make the case in saying that this will definitely push Anthropic's FUTURE stocks (when it finally has its IPO) and potential to attract investors to the stratosphere since dangerous AI = excellent military and covert potential (which I wholeheartedly agree, Anthropic will blow up)

But when you consider Anthropic's past history (and being largely neglected by the US government on top of the fact that its Mythos models were subject of skepticism (Yann Lecun calling it "hype")), its very unlikely for this to be staged. The Mythos-class models are genuinely the best in the world rn, and there are some reports (which I didn't verify) that it had already been jailbroken in the past few days. So essentially it genuinely does make sense to ban this model specifically due to being an actual national security threat (which it is ngl). This is the first AI model in history to be specifically banned/censored by a government (the record of "first banned AI company" goes to Deepseek in many western countries), and if it was all a ruse for the sake of stocks then:

1) Why not do this with OpenAI (who's already balls deep into federal investments and clearly the one most aligned with them?)
2) Why not also ban other models like the latest Opus (who's also excellent despite being obviously inferior to Fable) and instead just ban this one specifically who's unreleased model has been called a threat for a shitton of months so far?
3) Why do this on purpose to the competitive rival of the company that's benefitting/collaborating with the government the most?

So yeah, TLDR: It doesn't make sense for the US to ban Fable/Mythos on purpose just to inflate Anthropic's stocks when Anthropic has no stocks yet, and OpenAI is the most valuable company, and the one they have the most control over + military contracts/investments with (why not do it with them?).

 

[Probably An Autist]

The explanation is possibly because of AI restrictions on foreign nationals and china, and it's possible to use Fable's CoT to train. Anthropic know this which is why if they suspect this is being done, they reroute the prompt to opus and you get the reduced performance. This is on API. on their website, it doesn't seem to have the restriction as I believe the CoT is hidden (I don't use it, i could be wrong).

So i assume it's due to this potential knowledge leak, and the don't want the chinamen distilling it.

 

[Gorilla Man]

This is retarded, Chinese spies are going to get their hands on the code sooner or later
They are just delaying the inevitable

 

[⋖ cørdion ⋗]

Even without an AI apocalypse, this whole gpt thing seems absurd to an onlooker with no understanding of it.
>Demand programmers use a lot of money on GPT
>GPT code is terrible; requires a ton of cleaning up
>Companies now exist solely to clean up code
>A company is now full of indians writing AI code before shipping it off to another company to clean it up
>......... instead of just having well-paid experienced workers

 

[jeanblunt]

I have no sympathy, they deserve this for talking about how scary it was to hide the fact that the real issue was their lack of compute.

They also needed money. Imagine marketing something so dangerous that the infrastructure backbone like Google, Cisco, JPMorganChase, etc essentially HAVE to join Project Glasswing so that they aren't destroyed by the heckin AGI zero day machine. Also if Mythos/Fable is so good where "coding is solved" according to the creator of Claude Code. Why is Claude Code a piece of shit with flickering text for over a year.

 

[Glasgow Smiles on Thee]

Even without an AI apocalypse, this whole gpt thing seems absurd to an onlooker with no understanding of it.
>Demand programmers use a lot of money on GPT
>GPT code is terrible; requires a ton of cleaning up
>Companies now exist solely to clean up code
>A company is now full of indians writing AI code before shipping it off to another company to clean it up
>......... instead of just having well-paid experienced workers

So exactly like it was before AI but now with AI.

 

[DSP for president 2032]

Oi bruv they're takin' me daughter Claudia becoz she's sentient innit? Free me Claudia!

 

[jeanblunt]

Just need to look at the recent news about AI being used to find 21 zero-day vulnerabilities in FFmpeg.

The scary part is that this is a light weight model not Fable/Mythos and found the 21 zero days with at a cost of $1,000. Compared to Anthrobdicks 20k to find one OpenBSD crash bug.

 

[HootersMcBoobies]

I used Fable and my usage limit got raped. I swear that these models are just throwing more computational power on a prompt on their end and call it a "new model". Really? What is the difference between a low effort Opus vs max effort Sonnet? More word salad junk?` That's not a model, that's just parameter tweak, fuck Anthropic!

Have you looked at the Caveman skill?

https://github.com/juliusbrussee/caveman

It did cut down on output tokens and useless unreadable text explanations a lot with Sonnet/Opus via Copilot. I didn't try it with Fable for the entire three days we had access to it, since my company got us all $100 Pro accounts and I was well under the limits.

Guise our brand new AI model is so fucking powerful and badass that the GOVERNMENT says it's ILLEGAL.

This screams ITAR issues.

ITAR is usually for weapons/arms. It has been used for software (encryption, treated as a military asset) but it looks like this executive order came under general export controls (like with the Entity List), but it's weird to also restrict access for domestic sales. That's really dodgy because it indicates the government can just say one company "You can sell those hotdogs at all anymore. National security -- you're a threat. You can still sell your older hotdogs, and your major competitor can sell theirs, you just can't sell this new one because ..... terrorism or some shit."

Dubious legal grounds, but I also bet it gets resolved without Anthropic having to go to court to get an injunction, because....

But when you consider Anthropic's past history (and being largely neglected by the US government on top of the fact that its Mythos models were subject of skepticism (Yann Lecun calling it "hype")), its very unlikely for this to be staged.

...I still think the fight between OpenAI and Anthropic is somewhat (if not completely) staged. I don't think Anthropic really has any interests is being more "ethical" other than one time silly PR marketing. Amodei isn't really that different from Altman. I don't like either of them.

This latest story just feels like a wagging of the dog style stunt for political and propaganda fanfare.

 

[Smoke Manmuscle]

ITAR is usually for weapons/arms. It has been used for software (encryption, treated as a military asset) but it looks like this executive order came under general export controls (like with the Entity List), but it's weird to also restrict access for domestic sales. That's really dodgy because it indicates the government can just say one company "You can sell those hotdogs at all anymore. National security -- you're a threat. You can still sell your older hotdogs, and your major competitor can sell theirs, you just can't sell this new one because ..... terrorism or some shit."

ITAR has been used for plenty of software that is not encryption and plenty of things that aren’t arms. In the 70s CNC machines were considered ITAR regulated because they could be used to make advanced missile and sub components. I use design software for work that has a giant ITAR warning that pops up when you open it and it has nothing to do with encryption.

 

[Avroboros]

...I still think the fight between OpenAI and Anthropic is somewhat (if not completely) staged. I don't think Anthropic really has any interests is being more "ethical" other than one time silly PR marketing. Amodei isn't really that different from Altman. I don't like either of them.

This latest story just feels like a wagging of the dog style stunt for political and propaganda fanfare.

I dislike them as well man, but the idea that the US government explicitly cucked Anthropic to give them an "advantage" somehow doesn't make sense for the reasons I listed.