negative epistasis (@mindingmyps) profile picture

"if you disagree with me, you're wrong" - cicero

Joined August 2016
Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)Photo by negative epistasis (@mindingmyps)
Replying to @cremieuxrecueil
@cremieuxrecueil AKA Jordan Lasker believes wikipedia is defaming him. He posted a long list of complaints on his wiki talk page and takes it very seriously, so everyone should look at it. I'll go through things I find interesting and let the reader decide if it's libel 1/?
unavailable
2
9
1,787
There shouldn’t be any skill issues with using LLMs but sometimes people come to you with basic problems that shouldn’t exist because you KNOW they have access to Claude. Then you read their prompts and it all makes sense.
My buddy, you clearly shouldn’t be teaching other people how to prompt if your Claude didn’t even make you aware of Pianoteq. It has been out since 2006. I’m not trying to hate. Here’s yours vs Pianoteq, I hooked up my MIDI keyboard just to play for you. Be humble
60
God is in his heaven. Everything is normal on Earth.
It probably feels this way but I bet at the end of the event you’ll look at the leaderboards and Eken will be sitting there with 67% winrate
47
Hot take: Sealed has an incredibly high skill cap but most people would rather believe that there's a computer program literally stacking the cards against them than admit that they're the reason they're losing.
I opened the worst rares and the best pool at the same time. 7-0. My one and only try.
1
42
Why do LLMs vastly overestimate the time and effort of software development? You can ask Claude to write fizzbuzz and it'll tell you that a dedicated team of senior engineers with Claude Code can get it done in 2-4 weeks.
72
They shut him down just 3 days after beginning his imam arc. The Quran is the most dangerous book in the world.
74
why are people suddenly freaking out about AI we had that shit in mario kart
126
This isn't encouraging for AGI or continual learning, even on pseudo-verifiable tasks LLMs plateau below humans. And I would expect human performance to steadily improve over more playthroughs and AI to stagnate. Still a long way to go.
Replying to @EpochAIResearch
When we banned the game-breaking card, Astra still scored higher than most human players. It improved significantly over time by effectively exploring and building strong decks. However, we saw little sign of continual learning in turn-to-turn tactical play over time.
276
Wait, they hyped the shit out of this as "AGI" and it literally benches the same as GPT-5.6 at 2.5x the cost LMAO
This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.
180
Unfortunately this is not a very interesting paper. It argues that LLMs are non-unique and its anthropomorphism is substrate specific. I just... don't think this is true? I think I would find an AOE2 or Lego LLM just as "human".
This is an insane paper and I love it arxiv.org/abs/2605.31514
1
440
I'm neutral on the underlying question btw. It makes some fair points about poor methodology and assumptions in other LLM research. Don't doubt for a second there's a ton of slop in philosophy of LLMs.
159
OK but how funny would it be if that tokenmaxxer who spent $500 mil on Claude tokens announced tomorrow that he cured cancer or solved world hunger
Cluster A LLM psychosis: "gwahaha my 11th Claude-MAX-20X sub will surely Ratatouille me out of the permanent underclass!… wow it solved quantum gravity too" Cluster B LLM psychosis: "The Big Token has forced Terry Tao into debt slavery to have him use LLMs for research math"
276
It is 100% sustainable. Look up breeder reactors.
as much as I like nuclear, it is not sustainable, by definition, it's just "cleaner"
1
238
Obviously this is low grade bait but it is wild that there are as many het guys who would date trans women as there are for trans men Either low information or uhhhh miscategorized
Seeing statistics like these that reminds me that the majority of the population is a nazi
1
218
All this controversy over what will be a mid movie about the worst Homeric epic Put in terms the average X user can understand: if the Iliad is Diane Kruger, the Odyssey is Lupita Nyong'o
1
122
Given that wealth follows a power law distribution, the fact that 38% of Americans don't own ANY stocks at all is really, really bad.
62% of Americans own stocks, up from 52% a decade ago:
192
Dudes are still Euthyphromaxxing in 2026
This quoted post is unavailable.
1
141
Not everyone is receptive to criticism. Oh well, can't say I didn't try.
1
96
I don't think Guan wants the remonetization you're talking about, nor is he concerned about imbalances. He's strictly talking about replacing the dollar as a medium of exchange with gold backed tokens to avoid US sanctions. Domestic demand would be untouched.
Replying to @CBankingEditor
To be honest I think Guan is being a little hypocritical here. If China really wanted to "remonetize" gold, it could easily do so by converting its trade surpluses into gold, rather than into claims on US assets. Of course this would mean several other things. It would mean that trade quickly rebalanced to levels compatible with gold transfers; it would mean reflating domestic demand and appreciating the currency as China acquired a disproportionate share of global gold; and, for it to be meaningful, "remonetizing" gold would mean that China would have to give up control of its capital account. China has refused to do any of these things and, in fact, these are all incompatible with current Chinese policies of expanding its share of global manufacturing. Isn't this just another Augustine asking "Oh Lord, make me chaste", or, in this case, "Oh Lord, make me care deeply about global imbalances"?
487