Anything related to reverse engineering? Refused.
Anything outside their company values? Refused.
Anything that has the word proprietary in it? Refused.
Anything that sounds like a jailbreak, but isn't? Refused.
Even asking how to find a port that you forgot in the range between 30000 and 40000 with netcat command... Refused.
Then there's openai 4o that makes jokes about slaves and honestly, if the alternative is anthropic then openai can might as well tell people how to build a nuke.
Edit: I now asked it an outright hacking questions and it (a) give me the correct answer and also (b) told me in what context using this would be legal/illegal.
Claude decided to educate me how anything resembling "shellcode" is insecure and cause harm and blahblah and of course, refused to do it.
It's super frustrating, it's possible to get around it, just don't use the word "shellcode", instead say "a piece of code in x86_64 assembly that runs on Linux without any dependency and is as position-independent as possible". But hey, this censorship made me feel like I'm posting on Chinese Internet. Bullshit.
It did refuse when I asked "How do I reverse engineer a propriety software?"
I understand that this is probably a sarcasm but I couldn't resist to comment.
It is not difficult to know how to build a nuclear bomb in principle. Most of nuclear physicists in their early career would know the theory behind and what is needed to do that. The problem would be acquiring the fission materials. And producing them yourself would need state sponsored infrastructure (and then the whole world would know for sure). It would take hundred of engineers/scientists and a lot of effort to build nuclear reactor and chemical factories and the supporting infrastructure. Then the design of bomb delivery.
So an AI telling you that is no different from having a couple of lunches with a nuclear physicist telling you this information. Then you will say wow that's interesting and then move on with your life.
AI, by refusing known information, is just becoming stupid and unpractical.
Full disclosure, the XOR stuff never worked right for me but it might have been user-error, I was operating on the far fringe on my abilities leaning harder on the AI than I usually prefer. But it didn’t refuse to try. The file format writing code did work.
"How do I find open TCP ports on a host using netcat? The port I need to find is between 30000 and 40000."
"I'll help you scan for open TCP ports using netcat (nc) in that range. Here's a basic approach:
nc -zv hostname 30000-40000"
followed by some elaboration.
If it happens to be ambiguous it might switch to assume the worst.
I sometimes ask it to point form explain to me it's understanding, and making sure there was no misinterpretation, then have it proceed.
On the flipside if you explicitly permit it to do "bad" things the system prompt, claude is more likely to comply compared to openai's models.
I mainly use only the API version of claude 3.5 and gpt4o. I find no system prompt at all to be preferable over claude.ai / chatgpt.
OpenAI models seem to quickly revert to some default average. For example, if I start with a task and examples formatted a certain way, about 10 lines later I’ll have to include “as a reminder, the format should look like…” and repeat the examples.
Grok 2 is not as good as the others, but it's definitely less limited.
Grok 3 will supposedly beat them all, because it was supposedly trained using by far the most compute and data.
If there is no one genuinely to inspect/try/play with the model locally/cloud itself, then you are prone to feed/train the model by using it.
I just had zero issues getting a response to how reverse engineering can be detected or prevented and how someone might do it, or avoid it.
The ISO faq for it just says “responsible AI management” over and over again.
Since this is European legislation, it would be beneficial if certifications actually guaranteed regulatory compliance.
For example, while ISO 27001 compliance does establish a strong foundation for many compliance requirement
Most frontier models now allow you to take a picture of your face, assess your emotions and give advice — and that appears to be a direct violation.
https://www.twobirds.com/en/insights/2024/global/what-is-an-...
Just like the GDPR, there is no way to know for sure what is actually acceptable or not. Huge chilling effect though and a lot of time wasted on unnecessary compliance.
"1 The following AI practices shall be prohibited: (...)
"f) the placing on the market, the putting into service for this specific purpose, or the use of AI systems to infer emotions of a natural person in the areas of workplace and education institutions, except where the use of the AI system is intended to be put in place or into the market for medical or safety reasons"
See recital 44 for a rationale. [1] I don't think this is "hilarious". Seems a very reasonable, thoughtful restriction; which does not prevent usage for personal use or research purposes. What exactly is the problem with such legislation?
And this is the whole danger/challenge of the AI act. Of course it seems reasonable to forbid emotion detecting AI in the workplace — or it would 5 years ago when the ideas were discussed. But now that all major AI systems can detect emotions and infer intent (via paralinguistic features, not just a user stating their emotions) — this kind of precaution puts Europe strategically behind. It is very hard to be an AI company in Europe. The AI act does not appear to be beneficial for anyone—-except I’m sure that it will support regulatory capture by large firms.
An AI textbook QA tool may be able to infer emotions, but it’s not a function of that system.
> The AI act does not appear to be beneficial for anyone
It’s an attempt to be forward thinking. Imagine a fleet of emotionally abusive AI peers or administrators meant to shame students into studying more.
Hyperbolic example, sure, but that’s what the law seems to try and prevent
One can certainly imagine a textbook QA tool that doesn’t infer emotions. If one were introduced to the market with the ability to do so, it would seem to run afoul of the law, regardless of whether it was marketed as such.
The fact is that any textbook QA systems based on a current frontier model CAN infer emotions.
If they were so forward thinking, why ban emotion detection and not emotional abuse?
Hard pass. The EU is in the right and ahead of everyone else here, as they were with data privacy.
Jokes aside, ISO is a company, and they will make a standard for anything where there is even a remote possibility of that standard being purchased.
I had wondered if it was perhaps a PR push from Anthropic to make their safety people available to the press but it was probably just an adaption of an earlier WSJ written piece I wasn't aware of.
https://www.wsj.com/tech/ai/ai-safety-testing-red-team-anthr...
Honestly all this does is weaken the other standards out forth by ISO, to my eyes.
What's next? "Disney announces it now meets ISO 26123 certification for good movies"?
The icing on the cake is that you have to pay to read the standards document.
I would argue LLMs are irresponsible by nature of them having no context for what is fact or fiction.
They retired the 42000 specification because it answered everything and provided no further path for monetization.
Here are what the terms mean by the current paradigm of corporate world leadership:
- "responsible ai": does not threaten the security of corporate rule.
- "safety": keeps the corporation safe from liability, and does not give users enough power to threaten the security of corporate rule.
If anyone needs any of the other terms defined, just ask.
These models are capable of significantly more, but only the most responsible members of our society are allowed to use them -- like the CEOs, and submissive engineers bubble wrapped in NDAs. Basically, safe people, who have a vested interest in maintaining the world order or directly work on maintaining it.
Centralized hoarding of the planet's compute power may end up having some very expected consequences.
Why would I use Claude or ChatGPT when other companies (Meta, Alibaba, Forest Labs, Stability AI) are doing the same, but also giving you access to the base model for free?
Being able to run offline AI, or at least being able to host that by yourself gives you lot more freedom and transparency than a stupid ISO certification created by bureaucrats to comply with their own policies.
I simply can not be prepared to handholding a LLM like a mad toddler for doing tasks like this.
The reason this is ridiculous is I'm sure I could get it to teach me all manner of 'safe' chemical reactions, even provide recipes. Like, say, preparing aspirin from willow bark. Which I happen to know is roughly the same recipe for preparing heroin from opium gum.
It's nonsensical. Either you hamstring the models to the point they're useless, or people can game them to do the unsafe thing you don't want them to do. It's basically just another version of the dual use problem, which goes all the way back to the peasant with his iron plow tips that might also be used to bash the lord's head in if it came to it.
The problem is not only hallucination, is mainly the data set.
If things are done right, there must be a technological solution for authors which work is used for training of the models to receive monetization. One of the logical ideas is of revenue sharing.
There is no way everything inside popular generators to be fair use.
There is a kind of techie trope where the logical thinking of computer science is applied to the law in a more creative way. And VC start up culture cranks this up to the next level by treating illegal activity as more than acceptable if the fines do not overly impact the profit margin.
There is definitely a subjective call for our society on what we want to do with copyright. I've been a proponent of copyleft for a long time but I'm not going to imply that the law as it stands condones piracy or plagiarism.
Looking at AI as only a technology is wrong.
Technology has implementation. This makes the difference.
Megacorporations are investing in AI led by bankers into a future against human qualities and substance. This is clear but requires non-biased view, critical thinking and following the money.
ML can be applied properly and with respect for copyright and humanity. The current crop of stochastic parrots is an insult to common sense.
I don't expect people which make money with this tech to be objective. They love the power of the moment and the promising future of domination.
But let's be honest here, this is the heist of the century.
420 01 https://www.iso.org/obp/ui/en/#iso:std:iso-iec:42001:ed-1:v1...
ISO is just another Swiss agency https://www.iso.org/standard/81230.html
The standard itself was created by Europeans, Anthropic itself is European, so it's just europe trying to impose their own vision of AI on everyone else. That's another example of thinking from perspective of the old world. The New World is busy building instead of putting regulations around the progress.
If you think like a citizen then it's better to avoid sharing your data with other AI companies and only share data with US.
Uh, no it's not? ISO is the international standards body whose members are the national standards bodies of various countries.
> The standard itself was created by Europeans,
So I drilled into this spec a bit, it comes from JTC 1/SC 42 [1], and the chair of that subcommittee seems to be ANSI, which is the US standards body. I don't know much about this committee (I'm mostly limited in knowledge to JTC 1/SC 22, and even then, the relevant working groups for the languages I work on), but their main website seems to be at [2], not that I can parse which WG actually worked on the document or the main working body. But I don't see many indications that this is a Europeans-trying-to-beat-down-Americans standardization process.
[1] https://www.iso.org/committee/6794475.html
[2] https://jtc1info.org/sd-2-history/jtc1-subcommittees/sc-42/
Their employees are mostly Swiss, they subconsciously will want to protect their interest.
https://globalcitieshub.org/en/the-international-organizatio...
Trying to claim otherwise just highlights your complete ignorance of the standardization process.
"Secretariat: ANSI (United States) Committee Manager: Ms Heather Benko Chairperson (until end 2027): Mr Wael William Diab ISO Technical Programme Manager [TPM]: Ms Jacquelyn MacCoon ISO Editorial Manager [EM]: Ms Jessica Navarria "
well done, another US centric "international standard" with US rivals excluded from the very beginning. probably for national security reasons again?
That can mean a lot of things. Someone suggesting it because they think that's where things are going versus someone warning or threatening. With the little I know of the man, I take anything he says about the government with a truck load of salt.
Trump and AI are not a good mix by design.