general

I think the bubble popping and "AI causes human extinction" are sorta parallel to the capabilities of LLMs

or at least that's how I chose to write it, since there are a lot of permutations of combos otherwise

My take on LLMs is that it will never be super useful except for people who already know what they are doing, so the majority of people will be actively detrimental to them, or a sidegrade at best

LLMs will never reach human intelligence level, that is not what they are designed to do

human intelligence level meaning "it can actually think in a way humans do, not in the way LLMs determine what responses to give"

yeah, I was careful to not just write "AI is smarter than a human" or "better than a human at something"

because that's sorta meaningless

It also depends on if I am grading the AI doing math or not

just like how Quantum computers are "better" than traditional computers at one thing currently (simulating a quantum computer)

Not to rehash what I have talked about before too much, but AI has been really good at proving most of human "productivity" is a lie, and you could cut 60-70% of all (non physical labour) jobs and the world would probably still run

yep, and I still like your quote from last time a lot

I'll be curious if anyone else here is anywhere close to me on this survey

I live in the world where a lot of people around me are unironically the last answer on this, though probably without a hard stated deadline for that prediction

I think the problem with that is, is anyone who answers that is mentally retarded

I feel it to a degree, the idea that we are making a god

this mission has become religious for a lot of people

I mean, making a god in the literal sense of "deifying an entity" I can agree to

some people really seem to think these LLMs are alive

its interesting how many people fail the Turing test, almost completely unprompted

Akinator has a soul

sorta energy

and the other side of that extreme, is people who see a trendline going up forever, until literally reaching godhood status

I think a lot of it is a misunderstanding. I think entropy wise its something like 32 questions or something is the limit for what you need to narrow down anything in the universe from yes or no questions

but when Akinator does it for a small subset of things internet people would care about, they are amazed

As long as they make the ai’s fuckable, they can take my job

But i dont think thats happening in 5 years

We need japan to get on the ai train

I am really interested in when actual math models start getting use

the LLM brute forcing that accidentally proves some open questions is funny but actual axiom encoding structures is where the innovation will be

All the math stuff I have seen has been extremely guided experts walking a LLM through the process, aka, it is just a tool for ideas for the most part

I will be impressed with any of it if it can from start to finish prove something

I appreciate ideas being useful, but my personal experience with AI for math is it is wrong way more than it is right

there are some anecdotes of feeding a problem statement to a chatbot and waiting an hour

and then it accidentally gets it right

Yeah, if anything, any successes with AI and math feel more like crazy accidents then a big leap in technology

I think axiom encoding has promise as an idea for proving things, since really, you can already prove math using just formal logic and that is essentially what that is doing

but LLMs as they are, are a meme

I want to see an application of the fact that any math truth statement can be resolved as a solution to a Diophantine equation

this is a great lecture that mentions that https://youtu.be/2oD7VxEb17E

i did two doubles takes because i initially voted, saw 2030 and believed that was in 20 years, so i voted for another thing and then remembered 2030 is in 4 years...

I... I also did not clock 2030 as being in 4 years.

Wake up dan, its 2006, you are going to elementary school

yeah you would love to imagine an elementary school aged me wouldnt you

I literally met elementary school age you

not in person

Dan you must have hit your head dude, you got too excited when you heard about the nintendo wii coming out in a few months

2006 is a on the very lower bound

thats like kindergarten

True, I guess I would have been 7, possibly 8