You keep using that word. I do not think it means what you think it means.
I agree that LLMs do not understand what it is they are predicting. At least no more than any other computer program “understands” its output. Which is to say: they are not conscious beings experiencing things.
This doesn’t mean LLM output is random. It means they aren’t thinking persons. You may have confused random with non-deterministic. LLMs are complex guessing machines that require a fudge-factor to produce meaningfully useful output. If the guesses were fully random then they’d have a significantly narrower field of practical applications.
I just don’t understand why you think it needs to be “fully random” to be called random, when the output is in fact random. To me, when you can ask it the same question multiple times and receive different answers, it’s random and unreliable. You may argue it only changes its answer randomly 10% of the time or whatever- it’s still random. It doesn’t matter if it only answers incorrectly 10% of the time or 5% of the time. Every time the answer is unreliable, because sometimes it changes its answer for no reason.
That depends on your definition of “random”. They’re non-deterministic; the same inputs will produce multiple outputs, and I think that’s the point.
So, it depends whether you’re referring to most people’s definition of “random” or a mathematician’s, because they often differ. One characteristic of “true randomness” (as defined in mathematical terms) is that eventually, by accident, patterns emerge. When people inevitably notice these patterns, the system appears less random.
Systems that check ahead for such patterns and adjust outcomes to avoid them are called “pseudo-random”. Even though the outcomes are carefully calculated, they’re perceived as being more random because they lack patterns.
Notable examples I can think of off the top of my head are Diablo 3 and the official Risk mobile app.
In Diablo 3, the drop tables started off using a truly random number generator. Some players noticed long stretches of no legendary drops, other players saw multiple legendaries in a row. People complained the drops were “rigged”, so Blizzard altered the algorithm to be pseudo-random and players stopped ccomplaining. Now, if a certain amount of time goes by and you haven’t had a legendary drop, you’re guaranteed one. Once a legendary drops, you’re guaranteed NOT to have one for a certain amount of time.
With Risk, same story with the dice rolling algorithm; players abjectly refused to accept the outcomes were random, despite the devs adding functionality to count your dice rolls and every players’ showed a perfectly uniform distribution.
A mathematician would agree that an LLM isn’t random because the next node in the chain is determined by the current nodes. But, since the algorithm won’t generate the same output every time it’s provided identical input (like a calculator), that satisfies most people’s definition of the word “random”.
Bold claim considering there is no agreed upon definition on what we even mean by the term intelligence.
The ability to acquire, understand, and use knowledge.
the ability to learn or understand or to deal with new or trying situations
the ability to apply knowledge to manipulate one’s environment or to think abstractly as measured by objective criteria (such as tests)
the act of understanding
the ability to learn, understand, and make judgments or have opinions that are based on reason
It can be described as the ability to perceive or infer information; and to retain it as knowledge to be applied to adaptive behaviors within an environment or context.
Random? No. Probabilistic.
it’s inherently random, though, because it is working with things that have meanings and it cannot understand what the meanings are.
You keep using that word. I do not think it means what you think it means.
I agree that LLMs do not understand what it is they are predicting. At least no more than any other computer program “understands” its output. Which is to say: they are not conscious beings experiencing things.
This doesn’t mean LLM output is random. It means they aren’t thinking persons. You may have confused random with non-deterministic. LLMs are complex guessing machines that require a fudge-factor to produce meaningfully useful output. If the guesses were fully random then they’d have a significantly narrower field of practical applications.
I just don’t understand why you think it needs to be “fully random” to be called random, when the output is in fact random. To me, when you can ask it the same question multiple times and receive different answers, it’s random and unreliable. You may argue it only changes its answer randomly 10% of the time or whatever- it’s still random. It doesn’t matter if it only answers incorrectly 10% of the time or 5% of the time. Every time the answer is unreliable, because sometimes it changes its answer for no reason.
You don’t understand because you’re either not paying attention or because you’re trolling.
Random would be output like this:
Or this:
Random would be a password generator. Random would be utter nonsense. Incoherent. That’s not what LLMs produce.
Still not random. Also not “intelligence” at all.
That depends on your definition of “random”. They’re non-deterministic; the same inputs will produce multiple outputs, and I think that’s the point.
So, it depends whether you’re referring to most people’s definition of “random” or a mathematician’s, because they often differ. One characteristic of “true randomness” (as defined in mathematical terms) is that eventually, by accident, patterns emerge. When people inevitably notice these patterns, the system appears less random.
Systems that check ahead for such patterns and adjust outcomes to avoid them are called “pseudo-random”. Even though the outcomes are carefully calculated, they’re perceived as being more random because they lack patterns.
Notable examples I can think of off the top of my head are Diablo 3 and the official Risk mobile app.
In Diablo 3, the drop tables started off using a truly random number generator. Some players noticed long stretches of no legendary drops, other players saw multiple legendaries in a row. People complained the drops were “rigged”, so Blizzard altered the algorithm to be pseudo-random and players stopped ccomplaining. Now, if a certain amount of time goes by and you haven’t had a legendary drop, you’re guaranteed one. Once a legendary drops, you’re guaranteed NOT to have one for a certain amount of time.
With Risk, same story with the dice rolling algorithm; players abjectly refused to accept the outcomes were random, despite the devs adding functionality to count your dice rolls and every players’ showed a perfectly uniform distribution.
A mathematician would agree that an LLM isn’t random because the next node in the chain is determined by the current nodes. But, since the algorithm won’t generate the same output every time it’s provided identical input (like a calculator), that satisfies most people’s definition of the word “random”.
Bold claim considering there is no agreed upon definition on what we even mean by the term intelligence.