@ExoticLife3416@00000sol0@Kirsten3531 Yep. What is implicit to the framing of "AI solved an Erdos problem", as opposed to "a skilled mathematician used AI to solve an Erdos problem", is that a toddler should be able to get the same result out of the AI.
https://t.co/dTBua3VqtR
OpenAI’s latest “maths breakthrough” with LLMs doesn’t mean these models can actually solve math problems.
As anyone who actually uses AI knows: 99% of the magic is in the prompt engineering. You carefully lock the search space so the correct solution path is already baked into the context.
That’s why nobody “discovered” it before, the human did the real intellectual work upfront.
How many months has OpenAI been iterating on this single specific prompt? They’ll never tell you. But we already know the truth: leave you alone with raw ChatGPT and you won’t reach the same solution. That’s the lie. It was never the AI.
Besides the search someone still has to validate each candidate painstakingly. Making the whole process absurdly costly and inefficient.
Once the prompt is locked in, the LLM just samples from its training distribution like a blindfolded shooter spraying bullets. Fire enough shots and one will eventually hit. The rest miss. Different story.
This isn’t intelligence. It’s sophisticated brute force, think password-cracking algorithms on steroids. Wasteful, statistical, and fundamentally not intelligent but blind.
LLMs are incredible pattern matchers. Let’s stop pretending they’re junior mathematicians.
@realmikeai@tszzl Just 2 weeks ago I had Claude tell me that db request results in the backend I used were cached within the same db transaction, so it was safe to re-run the same request within a transaction. I built a whole backend with this in mind, only to find out it was false. Cool tool, thx
@realmikeai@tszzl The people that "get it" don't do any real work and therefore have no ability to judge the quality of the output. To this day I allow myself to be meme'd every few months into trying Codex or Opus with Claude Code or Pi agent with Qwen 3.6... and I stop using them within a week.
@nic_carter I still can't ask Claude to make me a million dollars out of nothing with no further instructions, and you dummies want me to believe this shit is 10,000 personal mega-geniuses at everything in a little box ready to slave away just for me.
@nic_carter Computers were already better than humans at an uncountable amount of things before AI. Was Google a savant genius to you? Was Shazam a virtuoso because it could identify songs better than any human from humming?
@00000sol0@ExoticLife3416@Kirsten3531 Closed problem surface != open problem surface. Brute-forcing known software attack vectors is more akin to playing Go and Chess than solving novel math problems.
@00000sol0@ExoticLife3416@Kirsten3531 Ofc the fact that anyone can whip up an AI that automatically searches for common security vulnerability has implications insofar as how much of the overall attack surface of software can be exploited now, but that's because it can attempt known attack vectors.
@00000sol0@ExoticLife3416@Kirsten3531 Same for security vulnerabilities btw. It has yet to find any security vulnerabilities that couldn't have been found by any run of the mill software security expert, but they are great at brute-forcing known common vulnerabilities in software.
@ExoticLife3416@00000sol0@Kirsten3531 Again, this is such a far cry from how people like you are trying to frame it where you pretend like you can just ask an AI "solve an Erdos problem" and it'll come out with 100 new ones solved in a couple days.
@ExoticLife3416@00000sol0@Kirsten3531 Erdős problem #1196 was solved by someone who had been trying to solve it for 7 years. He dumped all the relevant knowledge he accumulated around the problem in that time, and GPT 5.4 helped him create a formula that served as formal proof of what he was already able to describe.
where would $wojak be on this list without:
-crash shilled wojak on Eth vamping liquidity PLUS notable members of our community
-a wojak derivative launching and going to 25m
-everyone on CT pretending it didn't exist because of some notable holders (who are now out)
-endless, unsubstantiated FUD about the deployer, slurp, from innumerable accounts who clogged up the ticker for MONTHS
despite ALL OF THIS - excluding the two bundled pieces of meme dogshit on this list and pumpcade - wojak remains the 4th biggest memecoin on pumpfun.
-it is 85% LOWER than it's ATH
-it has been redistributing supply for almost half a year
-the dev has been routinely using creator fees to buy back over 4% of the entire supply and locked it into a multi-sig wallet with some of the most trusted core whales
-it is the most eponymous meme in all of the internet's entire internet history
can I make it any more clear to you what you are missing out on if you are not allocated?
@00000sol0@Kirsten3531 No, those Erdos problems had already been solved in niche papers that the people keeping track of Erdos problems weren't aware of. It's still a testament to the utility of LLMs to do wide context-aware searches on latent space but this is a far cry from them solving them.
somehow i made it to day 100 lol
it’s been an interesting journey making art for the most relevant and widely used meme in internet history
wojak’s versatility makes it fun, because there’s literally a $wojak for everything. it gives you a whole palette to play with, endless ammo for ideas, and that’s what makes creating around it so enjoyable. but at the same time, it gets tricky, because after years of people all over the world making wojaks and memes out of everything, not accidentally recreating something that already exists while still trying to be creative becomes part of the challenge (i guess that’s part of the price of being the most used meme on the internet)
thanks for the love, y’all supporting my illness made the grind way easier lmao
huge shoutout to everyone who shows up daily, especially @BOMO_io and @lvprism for being there every day. appreciate y’all
we keep goin till 6.9b