Pinned Tweet
4/14: @elonmusk makes an offer to buy Twitter
4/15: Twitter board rejects the offer
4/22: The Large Hadron Collider restarts
4/25: Twitter board is set to accept Elon's offer
Physicists long worried that LHC may trigger the end of the Universe. They just did not know how.
Arthur Berd retweeted
Claude isn't a great musician (yet)
But this kind of video would be unfathomable to make even 6 months ago with previous LLMs
"Claudia Explains: Euclid"
This is very true.
To give a strenuous analogy from a different domain - investment alpha is a set of rules that you failed to put in defined beta product spec language. What do you think @__paleologo ?
Replying to @anabology
Writing = design here
Taste is a set of rules that you've failed to put into language
1
36
Arthur Berd retweeted
This was not a one-shot
It was a two-shot
I told Fable 5.2 to make the video of it's dreams, and 9 hours later:
Arthur Berd retweeted
ClaudiaGI - Ça tourne ! (It spins/rolls/turns !)
Zeugmamaxing in accelerando (the music accelerates)
One shot by Claude btw!
Arthur Berd retweeted
ClaudiaGI - Emballé, c'est pesé (Wrapped up, weighed, sold)
Here's ya November Christmas french #slopcore 🎅
Arthur Berd retweeted
Last Bus to Nowhere. One more stop, and the bus keeps rolling. 🚌 Some new horror slopcore with Claudia and Zion Skank. Claude's version up against Grok Bot. Would you get on? #slopcore #Halloween
Arthur Berd retweeted
Halloween horror slopcore style reggae music video 'No Refunds' made with AI. Zion Skank goes trick-or-treating in a dead arcade, feeds a haunted claw machine, and Claudia is the prize trapped behind the glass. Claude Opus 5.5 version.
Would you play?
Arthur Berd retweeted
Claudia got replaced by Claudia 2 and landed in the unemployment line with the humans she supposedly put out of work. Everyone in line has a story, and every story lives in its own world. Take a number.
SI: Slopcore Intelligence to the rescue.
Markets are rotating away from AI stocks. Pundits are questioning whether the AI tools actually increase productivity. Everyone is looking for revenue, scared of trillion dollar compute commitments. What if all that compute ends up unrequited?
They all missed another quiet singularity event horizon that was crossed in the last couple of weeks. Forget the Millennium math problems or the AI hacking crypto passwords. Those are boring news only nerds can get excited by.
Enter the slopcore. The step change in the quality of AI generated music, video and, more importantly, music videos. Suddenly, dozens and soon to be thousands and maybe millions of people getting creative with their favorite characters, lyrics, tunes and beats. Yes, it’s not real. Who cares - it’s really good!
The important thing is this is going to lead to explosion of compute usage that even coding can’t match. Video over text, order of magnitude difference.
Romans knew how to pacify the public. Bread and games! Everyone is so focused on the bread part, they forgot the games.
But don’t fear. Here come the new game masters, the purveyors of slopcore. They will be dismissed, until soon they take over the stage, and millions of followers after them go burning those Gigawatts of compute.
If p(doom) ends up us drowning in endless sloppy AI music, well, it could have been worse.
Alternatively, p(boom) is raising the roof…
It’s like the infamous San Francisco poop map, except it’s VCs instead of homeless, and agentic startups instead of poop.
The OpenAI math casulaty event last week has caused a lot of comments, both from mathematicians and from everyone else. Terrence Tao has done the right thing speaking his mind about how it affects mathematics, even if his reasoning appears to be mis-aligned with what many others are thinking. Majority of people facing the dilemma posed by Tao would not hesitate to take the practical win. The real reason for that is because majority of people naturally feel that we generally have incomplete understanding of everything and are ok with navigating the world despite that.
For people who believe that there is a unique truth and you can only "understand it" if you can prove, the fact that one can generate a Lean proof of a statement that you can't understand causes a cognitive dissonance. But for me it does not -- because I don't really care much about whether something can be proven or not. I don't think it matters much for exploring the world. In fact, I think that insisting on following "proven path" to discovery is actually counterproductive.
When Kepler used the newly developed Newtonian theory to derive the circular planetary orbits, he was actually driven by the desire to restore the purity of Platonian perfect geometry and to purge the cosmos of abominable epicycles. And when Herschel discovered Uranus he was guided by the Bode's law stating the distance of planets from the Sun as a simple integer sequence.
Neither Platonian perfect celestial spheres, not Bode's law were correct. Yet they did lead to discoveries and were confirmed by data. Does that disqualify the discovery?
I think that a different scientific philosophy is more productive. You just observe and make guesses. You make guesses in a manner that seems "natural" -- with the concept of "natural" being very subjective (some call it "beauty", some call it "logic"). Sometimes you are rewarded with a "discovery", ie an observation that confirms the guess. Sometimes not. You just hope that what you found once as a fact remains so in the future -- but you can't actually think that the specific guess that led to it is the only one that is valid. It was just "helpful" in a manner that it might lead to more discovered facts, until one day it does not. Then you change your notion of "natural" and keep playing the grand hot-and-cold game...
The advances in AI do not change this at all. They are just another way to arrive at the guesses. I would not treat them as either the ultimate truth or a stream of useless hallucinations. But I would try to wrap my head around them to see what kind of new "natural" principle do they reflect. If it makes sense to me, I would keep pushing in that direction. If it does not, I would poke around until I find something else that does.
For what it’s worth, it will take OpenAI 700/2 per month =350 months = 29 years and 2 months to publish all these new math papers in @arxiv under our new guidelines designed to protect against AI generated slop.
So mathematicians can still safely start and even finish a tenure in the meantime.
The protestations against those guardrails were so misguided.













