🤣... this is truly my favourite piece of literature today. To read my frustration, anger and disappointment being so eloquently articulated has me curious... is it to be or not to be....
AI output is almost by definition the average of humanity … but that also means that it is better than roughly half of humanity can do .. the livelihood of knowledge workers in that half are being threatened … and as the LLMS get better that percentage will grow … I commend you to read Venkatesh Rao’s writings - particularly his ‘Peak Mediocrity’ themes …
Secondly .. if people below average are displaced by bots and AIs how will they earn money to survive - much less live … and who will have the money to buy all the stuff that the bots and AIs are producing?
Hm. I was thinking about that but I ended up not including this in the article.
I love these kinds of guesstimation games. I'll try to give an answer at the end.
My guesstimation is based on Mensa IQ. According to them, the top 1% is treated as "very superior" in terms of intelligence. That's about 80 million people.
However I read somewhere that skill variability does not plateau at all.
I can anecdotally testify to that. I have done the Mensa test and did qualify for the "top 1%" according to their criteria.
Yet even when I talk to people who I think are at least an order of magnitude smarter than me, I feel dumb as shit. I'd imagine Einstein level would need at least 2-3 orders of magnitude better.
So on that note, there are 80million people in the top 1%. 2-3 orders of magnitude would mean about 500k-1million. If intelligence distribution is random, then this sample is representative of the whole species.
I think when a child like that is born, affluence of parents will be a very strong determinant if they actually blossom into the geniuses they can be. Household income should be around 70-100k USD for them to be able to afford what they need to support the child into becoming the genius they can be.
This means about 30-70k people on Earth right now have Einstein quality IQ.
Now this is JUST IQ. A lot of things are not mentioned here. Einstein was a genius, Van Gogh was a different kind of genius. Add some margin of error and say:
About 100k people are alive today that are Einstein level genius.
Not all of them will actually focus on bettering humanity. Let's factor that in and say half of them will become some great contributors to our species.
Is the question, even from a technical view, really how much of it is genius-level output? Or is it sufficient that there is some genius-level output in there?
Then I'd also point out that the LLMs we are currently working with are trained, or fed rather, the data that was already available. If we look at why coding is something LLMs are so strong in, it is partially because almost everything there is to know about coding is and has been on the web, Stack Overflow, etc., and a second important factor is that coding languages are a special case of languages that are much more precise and strict in their “grammar” than human language is with all of its ambiguity and metaphor, etc.
Looking forward in time, current LLMs are going to create a huge amount of low quality content and data, but there are also people – I would argue Azeem being a great example – who are outputting higher quality content and data BECAUSE of and with the help of the technology. That data then is distilled or condensed and enhanced, and it will subsequently also be available as input (fodder) for future models that we grow (training is a bit of a misnomer, really).
Is there not a chance that we, or AI, will be able to separate the wheat from the chaff?
🤣... this is truly my favourite piece of literature today. To read my frustration, anger and disappointment being so eloquently articulated has me curious... is it to be or not to be....
haha thank you. once I finished writing I felt that this wasn't even an article. it was a rant with extra steps :D
Nope…it's mostly Cat Barf!
I'm trying to get famous for the declaration…
AI output is almost by definition the average of humanity … but that also means that it is better than roughly half of humanity can do .. the livelihood of knowledge workers in that half are being threatened … and as the LLMS get better that percentage will grow … I commend you to read Venkatesh Rao’s writings - particularly his ‘Peak Mediocrity’ themes …
Secondly .. if people below average are displaced by bots and AIs how will they earn money to survive - much less live … and who will have the money to buy all the stuff that the bots and AIs are producing?
What percentage of human output is Van Gogh or Einstein quality by your estimation?
Hm. I was thinking about that but I ended up not including this in the article.
I love these kinds of guesstimation games. I'll try to give an answer at the end.
My guesstimation is based on Mensa IQ. According to them, the top 1% is treated as "very superior" in terms of intelligence. That's about 80 million people.
However I read somewhere that skill variability does not plateau at all.
I can anecdotally testify to that. I have done the Mensa test and did qualify for the "top 1%" according to their criteria.
Yet even when I talk to people who I think are at least an order of magnitude smarter than me, I feel dumb as shit. I'd imagine Einstein level would need at least 2-3 orders of magnitude better.
So on that note, there are 80million people in the top 1%. 2-3 orders of magnitude would mean about 500k-1million. If intelligence distribution is random, then this sample is representative of the whole species.
I think when a child like that is born, affluence of parents will be a very strong determinant if they actually blossom into the geniuses they can be. Household income should be around 70-100k USD for them to be able to afford what they need to support the child into becoming the genius they can be.
This means about 30-70k people on Earth right now have Einstein quality IQ.
Now this is JUST IQ. A lot of things are not mentioned here. Einstein was a genius, Van Gogh was a different kind of genius. Add some margin of error and say:
About 100k people are alive today that are Einstein level genius.
Not all of them will actually focus on bettering humanity. Let's factor that in and say half of them will become some great contributors to our species.
That's 50k people.
So summing it all up, it's 0.000000625%.
Is the question, even from a technical view, really how much of it is genius-level output? Or is it sufficient that there is some genius-level output in there?
Then I'd also point out that the LLMs we are currently working with are trained, or fed rather, the data that was already available. If we look at why coding is something LLMs are so strong in, it is partially because almost everything there is to know about coding is and has been on the web, Stack Overflow, etc., and a second important factor is that coding languages are a special case of languages that are much more precise and strict in their “grammar” than human language is with all of its ambiguity and metaphor, etc.
Looking forward in time, current LLMs are going to create a huge amount of low quality content and data, but there are also people – I would argue Azeem being a great example – who are outputting higher quality content and data BECAUSE of and with the help of the technology. That data then is distilled or condensed and enhanced, and it will subsequently also be available as input (fodder) for future models that we grow (training is a bit of a misnomer, really).
Is there not a chance that we, or AI, will be able to separate the wheat from the chaff?