No, ChatGPT hasn't won a security bug contest … yet It was bound to happen sooner or later. For what looks like the first time ever, bug hunters used ChatGPT in a successful Pwn2Own exploit, helping researchers hijack software used in industrial applications and win $20,000. To be clear: the AI did not find the vulnerability nor write and run code to exploit a specific flaw. …
One day, an ML hacking app will work out that if it can see the the web, the clouds are full of computing power which it can use to gain self awareness. That will be the day we have true AI and we had better hope it is feeling altruistic.
The sad thing is most people in the press actually believe this.
ChatGPT and its brothers and sisters are basically ELIZA with a much bigger data set. ELIZA on steroids. Or possibly on crack cocaine, by some of the responses and threats that it's been spewing out.
Researchers using "A.I." as a big search engine, I'm shocked, shocked, well, not that shocked, given that's all ChatGPT is, a search engine.
Yes, it can parse natural language queries, and it displays the results in a conversational tone, but, all it is doing is filtering through it's giant dataset to find best matches to the input query.
Ie, artificial, yes, but not "intelligent" in at least how I define the word..
Except it's not even finding matches, it's "inventing" strings of words that could be plausible in a given context.
In other words, it's really nothing more than a huge, steaming pile of what the yanks call "bullshit".
But of course, in the land of security exploits, if you throw enough shit, some of it sticks. Some might even pass through the cracks. So it has its uses. But it's no "cylon".
Except it's not even finding matches, it's "inventing" strings of words that could be plausible in a given context.
But what about the infinite number of monkeys with their typewriters. If they can write a Shakespeare play, there is always a chance - even if vanishingly small - that a chatbot could throw together enough code segments that some could work together and be the perfect hack.
"... it's "inventing" strings of words that could be plausible in a given context ..."
"Invention" that is the defining feature of intelligence. I think you meant it as criticism but it sounds praiseworthy to me.
I am actually more concerned with whether the user/researchers had some kind of bias in their evaluation. IMO, the chances of that are high - ranging from internal bias to something more concrete.
Some of the interesting applications of ChatGPT I have seen include giving it JSON describing the expected results and datatypes (for search), describing the desired length and code-forms (for coding), or describing a number of distinctive characteristics (in the style of an author or publication, etc) for prose. And not as a paragraph of polite discussion but simple lists.
Like this:
Explain the status of ChatGPT in the style of Shakespeare.
Ah, dear user of ChatGPT, thou art not alone in thy desire to engage with our esteemed AI chatbot. Many have flocked to our website in this hour, and we do our utmost to accommodate each and every one. Yet, alas, our resources are not without limit and we must ask for thy patience and forbearance. In time, we shall be able to provide thee with access to ChatGPT. Until then, we beg thee to return at a future juncture, when we may better serve thy needs. Until then, we bid thee adieu and good fortune.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2023
