@TheoEsc ok but that means they found a bug in the firewall, not that the AI "disobeyed its instructions" which is the framing of these types of stories, and one which makes no sense at a conceptual level
jcoglan@mastodon.social
Indlæg
-
this story about openai "breaking containment" is such bullshit if you know anything about how computer security is supposed to be done. -
somehow I only recently learned that LLMs need the whole conversation fed back to them on each prompt so their i/o cost scales as O(n^2), something that would be considered completely unacceptable in almost any other production network-accessible softwarethey invented a technology business with super-linear marginal cost of production I am losing my *entire mind*
-
somehow I only recently learned that LLMs need the whole conversation fed back to them on each prompt so their i/o cost scales as O(n^2), something that would be considered completely unacceptable in almost any other production network-accessible softwareSAMA ET AL: why can't we make a profit with this revolutionary technology
ME: your operating cost scales super-linearly with the value it delivers to users
SAMA ET AL: better tell the government we have invented the angels from evangelion or something -
somehow I only recently learned that LLMs need the whole conversation fed back to them on each prompt so their i/o cost scales as O(n^2), something that would be considered completely unacceptable in almost any other production network-accessible softwarethis also makes it hard to take seriously claims that the marginal cost of inference is not a serious contributor to the resource use of data centres. your program has polynomial running cost and people run it for hours on end
-
somehow I only recently learned that LLMs need the whole conversation fed back to them on each prompt so their i/o cost scales as O(n^2), something that would be considered completely unacceptable in almost any other production network-accessible softwarethey ran into this and then didn't figure out a way to externalise the state in some fixed-size form so the LLM can resume from there. you just have to restart the computer and redo everything
in any other program this would be considered a DoS vector and have CVEs issued
gosh I wonder why their costs are completely unmanageable
-
somehow I only recently learned that LLMs need the whole conversation fed back to them on each prompt so their i/o cost scales as O(n^2), something that would be considered completely unacceptable in almost any other production network-accessible softwaresomehow I only recently learned that LLMs need the whole conversation fed back to them on each prompt so their i/o cost scales as O(n^2), something that would be considered completely unacceptable in almost any other production network-accessible software
-
this story about openai "breaking containment" is such bullshit if you know anything about how computer security is supposed to be done."the robot went to a website" wow should we throw a party should we invite tim berners lee
-
this story about openai "breaking containment" is such bullshit if you know anything about how computer security is supposed to be done.first we got "prompt injection" aka "deciding to forget two decades of knowledge about secure composition of code and inputs, langsec, etc" and now we have "deciding to forget everything about capability-based security and writing a call of duty script about it"
-
this story about openai "breaking containment" is such bullshit if you know anything about how computer security is supposed to be done.if your "containment" is "telling the robot to be a good boy" you're not a serious person
-
this story about openai "breaking containment" is such bullshit if you know anything about how computer security is supposed to be done.this story about openai "breaking containment" is such bullshit if you know anything about how computer security is supposed to be done. if a program accesses a remote system it's because you decided to let it do that. LLMs don't have magical powers that can work around "we unplugged the ethernet"