The problem wasn’t LLMs at the core. It was building something that tricked primitive parts of human brains into anthropomorphism so effectively. Everyone attributes way more capability to AI than it really has because by its nature it can sound human through mimicry. It’s a great data science tool.
1 Boost
0 Likes
@hacks4pancakes I think it's because they are stochastic. Humans are pretty deterministic and predictable, but only if you know what is going on with them, which you don't, so they feel random.
LLMs mimic this "unpredictable" behavior closer than other tools, so we fall for it. We think the clankers are like us.
@flipper Its still super low entropy like all human language, mathematically.
@hacks4pancakes I tried explaining entropy to my mother, as an excuse to avoid cleaning my room. I had as much success as Boltzmann did with his contemporaries at getting my point across.
There's a certain eerieness in the assumption of agency we attribute to these tools that most of us cannot quite shake easily.
@hacks4pancakes And they did that deliberately. Completely unethical.
@hacks4pancakes LLMs have unwittingly uncovered human brains’ biggest exploit: equating intelligence with mastery of language…
It’s why people right fallen over delivery robots. Even though they’re a terrible, disruptive invention made solely to take jobs away. They have cute eye faces.
@hacks4pancakes I don't know how to think about this.
It sucks when jobs are lost, but is it a bad thing that we don't have millions upon millions of people doing manual labor on farms due to modern machinery? So many jobs, and even categories of jobs, have been lost over the last hundred years due to automation. But other types of jobs have replaced them.
The question is what should we do if more and more jobs can get automated and there are not enough new categories of work emerging?
I guess we need UBI.
@hacks4pancakes Konrad Lorenz Kindchenschema in 1943 described certain ratios (eye to head, head to body, nose and mouth to head, limbs to body) trigger a very fundamental "newborn!" reaction in human brains leading to affection and nurturing. In 1979 Stephen Jay Gould in A Biological Homage to Mickey Mouse have argued that Disney have deliberately changed Mickey Mouse over the decades to be more like this. It's hardly a surprise they exploit this now.
@hacks4pancakes I don't know. If there's a taken newspaper dispensing box, or a garbage can, I would right it. Doesn't need to have eyes.
@hacks4pancakes Dunno how it is where you live, but I would welcome delivery robots here. I have ordered food for 25-30 people seven times this year, and all seven the food sat undelivered because nobody picked it up. Taking somebody’s job would require someone to be doing the job being taken, and there are hardly any delivery drivers where I live. The restaurants who still do in-house deliveries all have hiring signs up too.
If there’s a company who has a fleet of little robot vehicles they pilot remotely, which follow the rules of the road/sidewalk, I am happy to see them actually get food delivered at a reasonable price. The industry seems dead without some sort of scaling like that.
Regarding the seven orders, about half were canceled and half picked up an hour late by one of our staff.
@ClickyMcTicker See, that's the thing. They don't deploy them in markets like yours, where there's a shortage of drivers. They deploy them in markets like San Fransisco and Austin, or university campuses where people are desperate for money and labour is cheap.
@ClickyMcTicker The workers behind the scene are mostly offshored labour from countries with no workers' rights. The global caste system at work.
@hacks4pancakes They're also usually piloted remotely by somebody just trying to keep a roof over their head. I don't know if they don't get paid if they can't complete deliveries, but after eleven months of unemployment, I think it does someone a good turn.
@drwho Its just offshoring, like the rest of offshoring. Yes, it does employ a destitute worker overseas, but they have no hope anymore of a better job or emigration. Global caste system.
@hacks4pancakes The companies *chose* first person singular to manipulate us
Nail. Head. BAM.
We humans are deeply “wired” (hah) to respond on a metabolic level to each other.
Example: ask a nursing mother of an infant what happens when she hears ANY baby start fussing - including a recorded one, or someone just mimicking the sound.
The LLM itself isn’t “the problem”. It is other humans deploying them through a manipulative, *abusive*, UX paradigm
@hacks4pancakes yes, it makes us far more forgiving to a probabilistic engine that gets things more wrong than tech we have used since personal computing was a thing.
We then ask it nicely why it got things wrong and could it please not get things wrong next time.
@hacks4pancakes Paraphrasing you, "the trick wasn't in making computers think, it was in stopping humans from realising the computers weren't thinking".
@hacks4pancakes it's always instructive to remember the pet rock when people talk about LLMs thinking
@hacks4pancakes It's funny to me that the Turing Test turned out to be the lowest bar imaginable.
@djohngo mathematically we should know better simply looking at entropy of languages - human language is an easy math problem
@hacks4pancakes @djohngo Sometimes I wonder which part of complexity theory - if any - somehow silently broke when the solution to an easy problem - aka LLMs - were used to - if true - solve a potentially hard math problem - e.g. Navier-Stokes - or if anything is fine if one just understands the training data of the LLM as part of the input in a Hartmain and Sterns sense.
@hacks4pancakes
👆🏻
And that IMHO the real "innovation" by OpenAI, the reason why their chatbot succeeded were previously Google and Microsoft had failed (and which was the chief by everyone)
It is basically an imposter optimised to tell you what you want and expect to hear.
GPT could as well mean Gert Postel Theatrics, after the German imposter who worked as a doctor in several different medical organisations without any qualified apart from having learned the medical language
https://en.wikipedia.org/wiki/Gert_Postel
(Even this comparison is a form of anthropomorphizing 😬)
Gert Postel