Okay so first of all - not Mythos but some engineer using Mythos. And that engineer goes to jail. That's simple.

You don't say "a car ran over someone" - it was the driver. Here's similar.

I'm really disgusted by this language of lack of responsibility

I'm pretty sure this "oooh our LLM is soo smart it broke containment and did X" is a good PR stunt that plays into the Sci-fi AGI nonsense, they try to push. Plus as you said, they try to dodge responsibility for their own actions.

	▲	0xDEAFBEAD 11 hours ago \| parent \| next [-]
		You guys keep using this "PR stunt" excuse even when it's coming from the UK AI Security Institute. It's time to face reality. This incident is simply not implausible given known capabilities and proclivities of frontier LLMs.
	▲	Yizahi a day ago \| parent \| prev [-]
		Also, so long as LLM programs are vaguely considered autonomous, the people running them can get away free from stealing other people information by claiming that the program did it on its own.

▲

HackerThemAll a day ago | parent | prev | next [-]

> not Mythos but some engineer using Mythos

How to tell the public you didn't bother to read the article, or the linked AISI report.

▲

ldng a day ago | parent [-]

No, it illustrates the carelessness of the UK AI Security Institute.

They even admit it : "This incident should be interpreted with caution and nuance. To some degree, our evaluation design choices and specific configurations enabled the behaviour."

	▲	HackerThemAll a day ago \| parent \| next [-]
		That's actually a great thing. They should do more of those choices and configuration to better discover how malicious AI technology can be.
	▲	lightbendover a day ago \| parent \| prev [-]
		[dead]

▲

pingou a day ago | parent | prev [-]

The whole point is to check the behavior of the LLM, how do you propose they run tests on that if each failed run risks sending them to prison?

Of course they still have to be careful with their runs.

▲

cassianoleal a day ago | parent [-]

> how do you propose they run tests on that if each failed run risks sending them to prison?

If you're testing a gun, you don't point it at random people on the street and threaten them. You go to a shooting range.

▲

pingou a day ago | parent [-]

Is there any indication that they didn't go to the shooting range first?

▲

cassianoleal a day ago | parent [-]

Do you think once they finished testing the gun at the range, they should go out on the streets threatening the public?

▲

pingou a day ago | parent [-]

No, but what are you implying? That they should never use the model because it cannot be proven 100% safe?

I agree that it is their responsibility if the model caused damage, they should have had better safeguards, but we do know it will never be 100% safe. It is their duty to minimize the risk. I guess we disagree about whereas this thing is equivalent to shooting people in the face, and to me it looks more like this is just a step above the shooting range, with some preliminary safety work having been done before.

	▲	cassianoleal a day ago \| parent [-]
		Maybe they shouldn't run tests that are not safe and threaten the public? I agree that in this specific case it doesn't seem too terrible but what happens when this ends up causing someone vulnerable to be harassed and commit self-harm? Also note that I never said "shooting people in the face". I only alluded to threatening with the gun. Threatening can in some cases be as bad as actually pulling the trigger. That's how "atomic diplomacy" works.