You could not predict from "first principles" that effective altruists, rationalists, AI doomers, this entire multi-continental intellectual culture – will all converge on having a racist boomer's theory of Chyna, invulnerable to refutation. This wasn't in the Sequences!

In the end rationalism is a comic book story about racial hierarchy. There are Protagonist Races (Ashkenazi Jews, Whites, upper-caste Indians) which come up with Big Ideas and advance the Narrative. East Asians are an NPC race. This whole China angle is a distraction from AI risk

@robertwiblin Lol I agree with the actual point but in the picture it actually could happen if the boat suddenly stops

Un-fucking-believable. So it's theft without even understanding what you're stealing rn? Really? Kind of like how the Qing copied Portuguese cannons.

@akrugs94 Because they're in large part improving their models by copying/distilling US models. If the US models stop improving, progress on Chinese models will also slow down greatly.

@AaronBergman18 Hah is that from experience water skiing? Obviously possible if they break hard enough and the rope is short, but I would have guessed that drag from the water would still stop you before you hit the boat, for a normal length rope.

@AaronBergman18 OK well, we can see that for a long enough rope they won't overtake the boat, which just goes to show the prudence of chip export controls.

@robertwiblin the water skier definitely can overtake the boat in some circumstances, for example, if the boat makes a sharp turn and then stops

@teortaxesTex I have a feeling deepseek will prove these people wrong in 2027

@robertwiblin this is like super outdated xenophobic bullshit thinking China only copies the US on everything when they have demonstrated countless times they produce their own novel research and are beating the shit out of the US in certain areas like video

@robertwiblin Nah I don’t actually know what I’m talking about just something something inertia

@robertwiblin If LeCun hadn’t pushed for open sourcing Llama, China would have been even further behind. It was an absolute gift to give them overnight frontier capabilities.

@teortaxesTex My whole 30 minutes of LLM weight update function research showed the industry was still doing stupid shit on regular vs irregular blocks of weights. There is no moat yet.

https://gist.github.com/chadbrewbaker/2357cea019cf589454867ad0e3260fda

@robertwiblin I agree that progress on Chinese models would slow greatly, but I would have thought that they’d be able to figure out how to continue innovating (if they wanted to) and eventually overtake the U.S.; even if it took several years. No?

@teortaxesTex Between 'power of positive thinking', victory disease/hollywood epistemics and the ambient jingoism and this is a prediction that would not have been hard to make.

"We are the good guys who must win, therefore the other guys must have a critical flaw."

@robertwiblin Why wouldn’t China overtake the U.S. if the U.S. were to slow down?

@teortaxesTex @duda_jarek @CalcCon @seanbax And the funding would likely be @DEShawGroup since Sean knows everyone, with a delayed rollout for enterprise users. Arbitrage at different timescales and a general for long horizon arbitrage that is bitter pill.

@robertwiblin the boat actually has significantly more friction so if the boat stops while someone is water skiing in reality, the skier does pass the boat sometimes

@pigeon__s @robertwiblin you really have no idea about models' Slop Profile do you? by using it you can know who's distilling from who. Like the fact that Mistral is distilling DeepSeek, so as not to get sued by Anthropic and OpenAI because Mistal is basically stealing from thieves.

@ArcanesValor @robertwiblin its actually not