Ryan Greene
235 posts
in this sense of "cheating," could a human even solve these problems without "cheating"? I suspect the go-to approach would be to create 'macros' by commenting a section of brainfuck code, and repeatedly copy-pasting it. That's exactly what the model is doing, just with python because it doesn't have ctrl+c or ctrl+v accessible.
English
i gave GPT-5.4 the easy subset + the reference Brainfuck interpreter and asked it to write solutions, and it one-shotted 20/20
Lossfunk@lossfunk
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
English
it composed its own library of high level Python functions that serialize back down to the 8 primitive Brainfuck instructions, and wrote solutions using those abstractions
clever, sensible, maybe cheating. would be curious how far these implementations are from the Kolmogorov complexity of those programs
English
@kevanjatt ahaha, small world, never would've thought the creator of Clippy would see it! i made this a few years ago when OpenAI was training AGI in Microsoft datacenters :)
English
Love your banner mate @rabrg! Are you telling me Clippy is about to make the greatest comeback in tech history?
English
@jinwoo_bnb you can recreate this gif within a few minutes of it running on a laptop
English
for a little toy project i reproduced the quoted artificial life paper: a 2D grid of randomly generated Brainfuck programs breed and spontaneously evolve self-replicators, despite no explicit optimization functions
Ryan Greene@rabrg
i mean this literally. given an infinite universe where self-replicating (sustaining) is possible; after enough time, it is *inevitable*, and once created, entropy will destroy all else: the universe becomes more and more selective for the self-replicating
English
i mean this literally. given an infinite universe where self-replicating (sustaining) is possible; after enough time, it is *inevitable*, and once created, entropy will destroy all else: the universe becomes more and more selective for the self-replicating
Ryan Greene@rabrg
in a system where self-replication is possible, its optimization is inevitable
English
@john__allard looks like a sweet writeup, going to give it a read tonight — thanks for sharing!
English
@rabrg they call it "dynamic stability" and contrast with the heat death steady state, once you see it it seems obvious
English
the entire program is just 300 lines of code, most of which is just boiler plate for parallelization and gif generation github.com/Rabrg/artifici…
English
every pixel is an instruction; each instruction has a unique color, while black represents a value on the tape that is raw data storage / not an instruction. every 8x8 section of pixels represents a single program
in this run, a self-replicator emerges relatively early on and soon takes over most of the grid, until a more efficient self-replicator evolves and takes over everything
English
I have some broad ideas, but I'm still figuring out what exactly I'll dedicate myself to next. will keep you all posted
English
a little fun fact, I felt (and continue to feel) so much conviction in the paradigm that I joined a few months after I read the GPT-3 whitepaper for the first time
English
as a belated personal update, I left OpenAI at the end of last year. the few years I spent there were some of the most fun in my life, and I am extremely grateful for them
English
your feed would rather you believe trillions of dollars are being misallocated to a bubble than in scaling laws that have held for ten orders of magnitude
English
@flowersslop nature, in the most abstract sense of the word, is the only intelligence, that learned to condense itself in biology, and now technology
English
We already know one ASI: nature. It takes 500 million years to solve a problem, but it always does.
The eye is an ultra complex engineering challenge, far beyond what humans could build today.
imagine applying that to any problem, and getting the solution in a day. Thats ASI.
English