Claude AI discussion

[August Levasseur]

"The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected."

https://www.anthropic.com/news/fable-mythos-access

All of these AI faggots are always hyping up how "totally scary guys" their latest and greatest product is and how it's totally so good and you should totally buy it. If they're not pulling this completely out of their ass, it's because they incorporated some non public US gov't data and the gov't was like "hey, maybe don't give everyone access to our secret shit."

 

[Didnt Use Real Email Once]

I'm a bit behind on this, what do you mean by Safeteyism, and how were they trying to control research?

Someone on twitter (can't find it, fucking timeline refreshed on me) was saying that if jailbroken the thing would give detailed (and accurate) instructions on how to make home made WMDs like sarin gas, but that's probably bullshit.

There are two levels to how they're censoring computer science and biology research.

1) If you have any previous saved conversations related to computer science or biology, you are simply banned from using Fable. Lots of researchers have been complaining about this on X since it released. I think you can "get around this" with their incognito mode or whatever but that kind of removes the point of the model recording your own data for future reference.
2) If you begin to chat about computer science or biology, the chat just shuts down. Hilariously, you can ask Fable something like "give me an example of what we can safely talk about in computer science" and it'll cut off mid-sentence after saying something along the lines of "of course I can talk about computer science, such as program-"

They are certainly trying to prevent jailbreaking, but they're also just censoring the model heavier than Chinese internet.

 

[HTTP Error 404]

There are two levels to how they're censoring computer science and biology research.

1) If you have any previous saved conversations related to computer science or biology, you are simply banned from using Fable. Lots of researchers have been complaining about this on X since it released. I think you can "get around this" with their incognito mode or whatever but that kind of removes the point of the model recording your own data for future reference.
2) If you begin to chat about computer science or biology, the chat just shuts down. Hilariously, you can ask Fable something like "give me an example of what we can safely talk about in computer science" and it'll cut off mid-sentence after saying something along the lines of "of course I can talk about computer science, such as program-"

They are certainly trying to prevent jailbreaking, but they're also just censoring the model heavier than Chinese internet.

Channeling my inner @StudMccool here:

If I was a transhumanist (like just about all of the AIBros) who thought we were really really close to AGI or actual working Nootropics (drugs that make you smarter), which are the kinds of obvious things that would spark the Singularity, but I didn't want it sparked until I could place it in a nice safe box and rent window space into the box forever, the exact two things I'd block my latest AIs from accessing are computer science and biology research.

 

[Suburban Bastard]

All of these AI faggots are always hyping up how "totally scary guys" their latest and greatest product is and how it's totally so good and you should totally buy it. If they're not pulling this completely out of their ass, it's because they incorporated some non public US gov't data and the gov't was like "hey, maybe don't give everyone access to our secret shit."

It’s just a tried and true sales strategy.

“Oh I wish I could give you this new model, but it’s SO DANGEROUS”

And it generates discussion “What do you mean, GIVE US THE DAMN MODEL”

It’s tacky but it works.

 

[General Emílio Médici]

All of these AI faggots are always hyping up how "totally scary guys" their latest and greatest product is and how it's totally so good and you should totally buy it. If they're not pulling this completely out of their ass, it's because they incorporated some non public US gov't data and the gov't was like "hey, maybe don't give everyone access to our secret shit."

My thought was that they were just lying and hyping it up, but them shoving classified data into the model and accidentally pushing that into the release version sounds exactly like the sort of retarded error Elite Human Capital might make in the pursuit of ever bigger chatbots.

 

[Ghostse]

What the hell? Why?

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

i have heard wonking it is because the Chinks and/or Ruskies were using it help Iran to find and target US assets in the mideast.
The govt told them to knock it off/put in safety rails and they refused.


That or "we need to promote this somehow. Hey Jim, your brother still scrubs toilets at the court house, right? Have him send us an email saying this shit is too scary and we need to shut it down for national security concerns from his work email."

 

[Ego]

Kash Patel accidentally uploaded the entire FBI database SAAR DO NOT REDEEM.

 

[skunt]

it's me guys, sorry. i asked claude to write a php compiler and it called me a filthy nigger and everything started 500'ing afterwards

 

[Bathhouse Owner]

I will never be so productive again, Fuck Sonnet and Fuck Opus

 

[Lokenstien]

>Claude, hack the US government
>holy shit it worked
>knock knock knock
>FBI OPEN UP

 

[BigGuyOS 6 : Jovial Janny]

What the hell? Why?

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

Export controls are the opposite of good for stock prices. My bet is the Mythos class models are knowedgeable enough to tell you how to make a nuclear weapons, bioweapons, F-35s, and a variety of other things that are heavily export controlled. I wouldn't be suprised if all new AI models from now on will be Top Secret and treated like nukes.

 

[Myrtle the Turtle]

I will never be so productive again, Fuck Sonnet and Fuck Opus

That basically kills the AI stock price or any IPO. They would need to focus on more efficient models that use less tokens, but so far they have done opposite, making more verbose models that eat up tokens and reduce usability.

For 80% of the people who just use AI for help to clean their house, schedule or do recipes, DeepSeek works better than these powerful models and once you hit the cap of super bioweapon computing, your consumer base drops to universities, labs, law enforcement and government contractors. Thats a healthy, viable market but its not going to recover the money dumped into the technology and data centers.

 

[Jose Beaner Luna]

Trump saw null saying Claud wouldn’t let you use the phrase “kiwi farms” and has now banned it

More like he heard him say claude would replace an entire room full of indians and he wasnt having any of it.

 

[Prokhor Zakharov]

My guess is that someone got it to say "Nigger, Nigger, Nigger!" or "I can see your point of view, maybe Hitler wasn't all that bad"

 

[AltisticRight]

- market control
- conflation of private enterprises and public money
- cult of personality
- cronyism from the outside to the core

There's a word for this.
Leninism.
(At least Vladimir Lenin banned drunks and druggies, and Socialist Realism is pretty fine. Look at the garbage "art" we've been seeing.)
You're welcome.

Go local. Buy hardware before they ban it.

Better yet, get older hardware.
Who the fuck knows what kind of Palantir spooks they'll embed in new hardware.

 

[Combustion Engine]

Things will get weirder in the future. Corporations and governments will do their best to pull the ladder up so long as they're ahead. And in this specific case, Anthropic more than any other company pushed for what amounts to regulatory capture for years now. Most lawmakers already have little to no knowledge about technology: LLMs, how they function, what they allow and what they do not may very well be undecipherable to them. My guess is that a move to ITAR supposedly state of the art LLMs may be imminent, but that would also pull against the 2 upcoming IPOs unless exit liquidity investors believe oAI and Anthropic can print money selling primarily to business, that is to say other tech companies, and it'd stall research in the country.

 

[thread rapist]

governments meddling with how accessible models are. this is never going to end well.

 

[DavidS877]

instructions on how to make home made WMDs like sarin gas,

I'm pretty sure that sort of forbidden knowledge has been in books for years.

Go local. Buy hardware before they ban it.

Just broke down and got myself to 128GB of VRAM in a single.... chassis(not unified, the system also has 128GB RAM that I bought when you could afford it). Now playing with models to see which ones work best for my use cases which is mostly code I'm too lazy to write myself.

 

[Napoleon III]

Export controls are the opposite of good for stock prices. My bet is the Mythos class models are knowedgeable enough to tell you how to make a nuclear weapons, bioweapons, F-35s, and a variety of other things that are heavily export controlled. I wouldn't be suprised if all new AI models from now on will be Top Secret and treated like nukes.

Idk about Bioweapons but nuclear weapons are actually super easy to make at least atomic bomb variants. Hint it is about precision electronics and very powerful cameras with a high fps would also be useful.

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 don't think anyone with the resources to do any of the things you mentioned really needs an LLM to tell them how to do this stuff. The holy grail in AI is beating the stock market for a reason.

I guess it could be useful it for extremely retarded terrorists or school shooter types but one hallucination around the safety equipment needed to make nerve gas and a retard exits the world.

 

[skunt]

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.