WormGPT Is a ChatGPT Alternative With ‘No Ethical Boundaries or Limitations’::undefined
I cannot possibly see how this could be a good thing.
Did you check out the article, because it’s most definitely not a good thing. It was created to assist with cybercrime things, like writing malware, crafting emails for phishing attacks. The maker is selling access with a monthly fee to criminals to use it. This was unavoidable though, can’t put the tooth paste back into the tube on this one.
Good point and all, but my first thought was that it could finally tell me who would win in various hypothetical fights lol
Wasn’t that a show on Discovery at one point? Deadliest Warrior. It was simulations using different technologies to figure out who or what would win in a fight. Newer technology would certainly make it more interesting, but you can only make up so much information, lol.
I work in Cybersecurity for an F100 and we’ve been war gaming for shit like this for a while. There are just so many unethical uses for the current gen of AI tools like this one, and it keeps me up at night thinking about the future iterations of them to be honest.
Treat CVEs as prompts and introduce target fingerprinting to expose CVEs. Gets you one step closer to script kidding red team ops. Not quite, but it would be fun if it could do the network part too and chain responses back into the prompt for further assessment.
We’re expecting multiple AI agents to be working concert on different parts of a theoretical attack, and you nailed it with thinking about the networking piece. While a lot of aspects of a cyber attack tend to evolve with time and technical change, the network piece tends to be more “sturdy” than others and because of this it is believed that extremely competent network intrusion capabilities will be developed and deployed by a specialized AI.
I think we’ll be seeing the development of AI’s that specialize in malware payloads, working with one’s that have social engineering capabilities and ones with network penetration specializations, etc…all operating at a much greater competency than their human counterparts (or just in much greater numbers than humans with similar capabilities) soon.
I’m not really even sure what will be effective in countering them either? AI-powered defense I guess but still feel like that favors the attacker in the end.
Everyone talking about this being used for hacking, I just want it to write me code to inject into running processes for completely legal reasons but it always assumes I’m trying to be malicious. 😭
I was using chatGPT to design up a human/computer interface to allow stoners to control a lightshow. The goal was to collect data to train an AI to make the light show “trippier”.
It started complaining about using untested technology to alter people’s mental state, and how experimentation on people wasn’t ethical.
I’m sure you were joking, but try https://www.jailbreakchat.com/
Not joking actually. Problem with jailbreak prompts is that they can result in your account catching a ban. I’ve already had one banned, actually. And eventually you can no longer use your phone number to create a new account.
As more people post ai generated content online, then future ai will inevitably be trained on ai generated stuff and basically implode (inbreeding kind of thing).
At least that’s what I’m hoping for
the thing is, each ai is usually trained from scratch. There isn’t any easy way to reuse the old weights. So the primary training has been done… for the existing models. Future models are not affected by how current ones were trained. They will either have to figure out how to keep ai content out of their datasets, or they would have to stick to current “untainted” datasets.
there isn’t any easy way to reuse old weights
There is! As long as the model structure doesn’t change, you can reuse the old weights and finetune the model for your desired task. You can also train smaller models based on larger models in a process called “knowledge distillation”. But you’re right: Newer, larger models need to be trained from scratch (as of right now)
But even then it’s not really a problem to keep ai data out of a dataset. As you said: You can just take an earlier version of the data. As someone else suggested you can also add new data that is being curated by humans. If inbreeding actually ever happens remains to be seen ofc. There will be a point in time where we won’t train machines to be like humans anymore, but rather to be whatever is most helpful to a human. And if that incorporates training on other AI data, well then that’s that. Stanford’s Alpaca already showed how ressource effective it can be to fine-tune on other AI data.
The future is uncertain but I don’t think that AI models will just collapse like that
tl;dr beep boop
Oh goody the AI hacker wars are just around the corner!
A scary possibility with AI malware would be a virus that monitors the internet for news articles about itself and modifies its code based on that. Instead of needing to contact a command and control server for the malware author to change its behavior, each agent could independently and automatically change its strategy to evade security researchers.
to quote something I just saw earlier:
I was having a good day, we were all having a good day…
now… no sleep. thanks
The limiting factor is pre existing information. It’s great at retrieving obscure information and even remixing it, but it can’t really imagine totally new things. Plus white hats would also have LLMs to find vulnerabilities. I think it’s easier to detect vulnerabilities based on known existing techniques than it is to invent totally new techniques.
