Whenever I have a question about science or math I ask a LLM to explain it to me with sources. It just takes less time than trying to find the sources myself and if I enter into the prompt that I don't understand certain parts it dumbs it down for my retarded brain. If I do this as a 30yo man isn't this just going to be how everyone does it in the next generation? Just like every one uses search engines now, everyone is just going to learn all their science from LLMs. Eventually even in higher learning people will just ask for their school's locally run LLM to design a research model and whatever on a certain budget. And eventually they'll just ask it to come up with hypotheses. Is it over?

>>16700525 (OP)
I use LLM's (Grok, Gemini, etc.) as much, sometimes more than legacy browsers such as Google and I'm 22. It's already a new era.

>>16700525 (OP)
Yes, and this is the "worst" it will ever be. Next year's model, hell even next week's model will be better and it will only get continuously better and better, cheaper and cheaper.

>>16700525 (OP)
How often do you verify the correctness of its answer. Be honest.

>>16700525 (OP)
No. You're a retard asking questions you don't understand of a machine that doesn't understand anything, and thinking you're smart for doing so.

>>16700525 (OP)
Most of the time (math) it gives you the proofs/definitions wrong (unless they are trivial) so you still have to figure things out. It even recommends me books that don't exist. Idk how it does for other sciences though.

Whenever I have a question about math, which happens a lot when writing AI models, I break it down into parts and take the time to study the fundamental concepts. If I offloaded my thinking to an LLM then I would not fully grasp the concept. Am a 29 year old man.

>>16700578

Things that writing neural networks for machine learning has taught me, coming from a background of calculus:

>Backpropagation
>Sigmoid function
>Partial derivatives
>Mean squared error
>Stochastic gradient descent
>Multidimensional linear regression

I will sit down and work through every part of the problem until I can wrap my head around it completely.

I especially like how partial derivatives are used to fine tune neural networks to local minima in functions that include thousands of dimensions.

What You Asked is Partial Co-Opetition, More Collaboration. Science is only slightly Used for It's over.

>>16700525 (OP)
you said science and math but consider coding.
now when i write code it feels like cheating. it is so fucking easy it is almost unfair to ask to get paid for it. the only thing i don't get is how are new startups surviving? anyone could write any kind of code now, the question is how are you going to sell your product? what new products do we even need at this point? we have cloud and we have AI. we can create our own "products" or just ask AI to do what you need directly.

>>16700525 (OP)
Search engines have just become worse. If you make it into an advanced search engine with good sources it will be the best. I want a search engine that lets me search through all videos and pictures in the world for any obscure thing like everyone who wears a gold bra and has a certain cup size or everyone who wears a specific type of watch, or everyone who said "holy bageezer". And then if we're smart instead of dumb we could make it into a networking and cooperation tool like a sort of wikipedia of everyone's ideas that intercompares and sorts out all information in the world such that always the guy with the smartest idea is the one that's the world wide known best leading to an extreme growth in research speed and quality everywhere.
etc.

>>16700697
I don't think LLMs have made any progress at making better code. Call me back when they can take the code that generates their models and make it 90% more efficient.

>>16700525 (OP)
how is using an llm different than a calculator?

>>16700908
If you used LLM's two and a half years ago to make code you know they could barely make a for loop. Now you can use them to cut your work by easily half at least (I can attest to this). I appreciate the input, but please understand the abilities of something before you judge them.

>>16700879
That's a pretty cool idea actually lol

>>16701197
Yep. I regularly visit my old code and ask for optimizations/improvements to the code. Its great at refactoring too

>>16700566
are you not aware that to train these AI models there literally isn't enough electricity to improve them much further? and transistor size can only get so small, so, barring some incredible sci fi tier breakthrough in various areas of tech simultaneously, we are rapidly approaching the physical limits of the universe with regards to compute

>>16701222
Just build more.

AI is already poisoning itself. There were experiments where they tried to expand LLMs by feeding them synthetic content generated by LLMs. It turned to complete shit.
Now when they scrape the web, they are ingesting synthetic content, unable to filter it out, and it's damaging their models. Going forward, all "new" content will be suspect. Even published material in books, magazines, and other physical media is now being poisoned by LLM generated content. To expand, they'll have to scrounge the libraries of the world, digitizing anything they can find that hasn't already been digitized. Much of this will be of limited value and it won't take long for it to run out.
Due to LLM poison, LLMs might actually now be near their apogee.

>>16701238
For as long as our cutting-edge is LLM's (glorified Markov Chains), this will always be an issue. However, I bet in the next few years we will see AI models that are more reliable in actually generating novel insights so this isn't an issue in the longrun.

>>16701238
>referring to that nonsensical "experiment"
yikes

>>16701222
"There's not enough electricity!!!"

Ok make more

>>16700525 (OP)
Extreme divides will form between those who think for themselves and those who plagiarize everything. It will show us all who is real and who is fake.

>>16700525 (OP)
You just use Wikipedia.

You shouldn't use a retarded program that search google for you, steal random data and then rewrite them for you

>>16701227
>>16701274
That would require massive investment in nuclear power and the oil barons will never allow that

>>16700578
Nothing wrong with asking questions or having a teacher to learn. You still need to internalize it as you say, but it can help. You're not 'offloading thinking'.

>>16701197
>cut your work by easily half at least
How do I learn to do this? The results I get are pretty useless.

>>16701265
Sounds like religious faith.