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 software
-
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 software
-
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 software
@jcoglan i was thinking about this, and i believe that simple inference is O(n^2), because it is matrix multiplication of input vector (input tokens). The situation with refeeding the whole context, which also has hiddeng growth, or even “agentic” approach - launching secondary queries by agents - makes the situation far worse.
My take is that the general strategy of “AI” companies is - in principle - throwing exponential complexity brute force on problems and showing off that sometimes it gives interesting result. -
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 software
they 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
-
they 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
this 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
-
this 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
SAMA 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 -
SAMA 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 somethingthey 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 software
@jcoglan @buherator This also means you can “gaslight” the LLM by manipulating the conversation before it’s sent back to the LLM
-
they 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
@jcoglan Have-You-Tried-Turning-It-Off-And-On-Again as a service
-
@jcoglan @buherator This also means you can “gaslight” the LLM by manipulating the conversation before it’s sent back to the LLM
@schrotthaufen @jcoglan And coding agents gaslight the user by injecting "system reminders" prefixes that are invisible to the user but the LLM sees them as user messages and regularly refer back to those as user requirements. -
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 software
@jcoglan Sort of, but not really in reality. What happens is that you keep the context cached in VRAM and only transfer what's new.
It's like complaining on GIMP needing to have the whole image in RAM to be able to work on it.
-
A abekonge@venner.network shared this topic
