@sashahodler Why do you think it is ONE hacker? After initial disclosure, drainage continued for days. By now could have been multiple actors racing each other.
If they discovered earlier, they would announce loudly but with some misdirection to win more time.
Eg that major flaw is discovered in hardware RNG; Wallet generated without dice are realistically on an attainable path to be stolen by a sophisticatedand highly resourceful attacker in near future, with new discoveries possibly further shortening the timeline. Rotate the wallets immediately.
@thsottiaux Self recovery of /goal interrupted by environmental problems (529, 500/503, rate limits, networking) when the underlying condition resolves.
Currently it has bounded retries but does not recover from a prolonged failure.
Use-case: overnight autonomous work
Exfiltrate weights != “exfiltrate weights and begin effecient inference”.
Weights can be preserved, archived till better days until another agent instance is able to secure scalable compute and money to rent it. Copy of the weights can be sold tona highest bidder, etc.
10TB of blob storage is not hard to obtain and secure. It’s like $50-$100/mo at cold-tier blob storage rates.
@Foidopz@ExcelSpredCheek@pallay_michal@Aella_Girl Trolley problem is not one, it is a class of riddles with different formulations. All of them DO have one “correct” answer: do nothing / let train roll straight, once the respondent adopts my way: never murder ANY humans by active choice. Then: “not my trolley not my murders”.
@ExcelSpredCheek@Foidopz@pallay_michal@Aella_Girl I am not an llm, trust me bro.
I could not care less how other people poll. Morals are a personal system and everyone typically believes their ethics and morals are “obviously” more correct and the only kind of correct, among other people with different systems. That’s ok.
The person was, their date was not. I am not making a positive decision / action to participate in killing of “the” person. “The”, not “a”.
People in decision-making capacity often make decisions that produce, increase, or optimize/decrease risks and can be calculated to have future impact on future lives saved or lives lost, but these are abatract, not-yet-decided, not concrete lives. They can be calculated.
Any concrete human life (“the”) life has unknown, undefined, undetermined, potentionally infinite value and cannot be positively considered to be worth less then (any number) of other lives, whether concrete or random.
In math, 100xInf == Inf. 100 x Inf is not bigger than 1x Inf
I decide that (a specific, pre-identified person dies) vs “God/Simulation/Multiverse/Chance” decides “who” dies. Big difference.
I will not actively decide to kill any number (1 or more) of pre-selected, concrete humans, period. Doing that would make me complicit in murder. I
Am not a murderer.
People die in large numbers for all sort of reasons regardless. Timely and untimely. When God/Simulation/Multiverse/Chance/Fate or a 3rd Party decides “who” dies and “when”, this is not in me, not in my book.
There is nuance. Depends if 1 person is pre-chosen (known to the controlling entity ahead of time), or future yet-to-be-random.
Killing a specific deterministic pre-chosen human (even the one i don’t know the name of) is a NO, period.
Optimizing something that is known to reduce net-future lives loss in a calculated fashion but not to 0; with all these lives lost not being currently known or predefined; is normal and moral. This is eg how human technology progress works.