Assuming you don't want the standard protections on user-uploaded content to apply for whatever reason, it's far easier to just throw user-uploaded images into one of the billion cheap little convnets trained to detect illicit material than it is to demand your users, who come to you solely for an anonymous platform, dox themselves.

> given that the trajectory for all models is a kind of singular “peak” performance.

Not sure I understand this. Most other AI researchers I've spoken to don't expect an arbitrary hard cap on model intelligence, and those that do don't think we're close to it yet.

As a concrete example, ARC-AGI 3 is great - the best models can't reach 10 percent, but even a negro could get something in the 90's, or at least the 70's.

> Deepseek V4 beats closed source models brutally based on any metric. And it does fine on math/coding/agentic/etc.

Hard to find recent metrics, but it does look very promising so far, and I love that it's open source. Lots of fun things can be done with that.

Anecdotally, I haven't seen the panic I saw during DeepSeek's original release. CAISI claims it's benchmaxxed and performs worse (GPT-5 level) on non-public benchmarks. They might be lying, but sentiment is that they aren't.

A more critical take than mine is here: https://thezvi.substack.com/p/ai-166-google-sells-out?open=false#%C2%A7seeking-deeply

> I would also say that the open source model is simply more sustainable long term in any competitive environment. Linux dominates almost all infrastructure

Open source's value to Linux is that every autist on Earth can scour its source code for problems then share out the fixes, and most do. The same advantage doesn't exist for model weights, which can't be patched by Sergei the160 IQ Siberian autist operating from a 1995 Compaq machine. There is an advantage to being the universally-used research platform, but it's not direct nor as large in magnitude.

> As far as “America is behind because of Indians,” this is a kind of “pigs can’t fly because they don’t have wings” argument that I do not understand.

If we once had a great reserve of flying pigs, and we systematically clipped the wings off of almost all of them, but kept a handful of unclipped ones in a lab somewhere, those pigs would be of note to anyone in need of air-mobile pork.

To be clear, jeets have clogged up the recruitment and education pipeline in a way that has downstream effects even on the high-importance research teams that are allowed to not hire them directly. But a team that does not hire them will do much, much better than the U.S. tech industry's current state would suggest.

For the record, I am rooting for China here, on the basis that I don't trust any of the U.S. frontier companies.

I use Brave, but use Ublock for ad-blocking. Much more versatile, and lets me manually nuke annoying website elements that don't technically count as ads.

If the Chinese want to read my 10 page screeds about the Jews, then they're welcome to do so. Maybe they'll learn something.

If Mark Levin actually wanted to alleviate anti-semitism he would publicly apologize to White people and then livestream his suicide by self-immolation.

> Further, I will continue to maintain that Chinese AI is much better and is actually driving development now, rather than the reverse. Qwen and Deepseek/Huawei (Qwen also integrating Huawei) are the defining AI models already, and the question now is the effectiveness with which US companies will be able to rip them off, particularly with relation to efficiency.

That's an extraordinary claim. The benchmarks seem pretty unbiased, and they seem to favor the American models.

As other commenters have mentioned, jeets are slowing America down with their incompetence, but the teams working on the important bits are almost all White or East Asian.

I've long said that it has enormous shitposting potential. The only suno song I've seen catch on to any degree is this one, though:

https://suno.com/song/45a7a5b4-7c47-4958-afd7-7f14282e2fea

My theory is that it reads as low-effort OC and that audio alone isn't a great medium for spreading over the internet, but with well-written lyrics these songs can be extremely funny. Once AI can be used to readily make similarly funny music videos, I expect it'll take off.

> AI can be used for war, because the Jews don’t care about civilian casualty rates.

But that's not the issue. The problem isn't that it does a good job of something-or-other but also kills a bunch of civilians. The problem is that it's roughly as effective at targeting as a headless chicken.

It's not bringing anything to the table, as far as I can tell. The U.S. was posting video of missile strikes on decoys and paintings, which I don't think would be happening if AI was providing useful targeting assistance.

It's about the Ukraine. Kiev's core strategy is finding freaks like this and convincing them to commit terrorist attacks. They do the same thing in Russia.

This isn't even the first time it's happened. Ryan Routh was, according to mainstream news sites, in contact with a "Russian speaker" who said he'd be able to live a normal life in the Ukraine after the assassination.

I suspect it will be determined by two competing doctrines.

+ We see the ZOG doctrine employed by Israel and the Ukraine, which involves gathering up third world fodder and trying to use Western command and control to succeed where Robert MacNamera failed. The intent is that AI will allow for fine-grained micromanagement of forces that obsoletes the need for intelligent, agentic soldiers.

+ We also see the anti-ZOG doctrine employed by Russia and Iran, in which units of bold, agentic men work in concert with FPV drones, supported by an inexhaustible supply of long-ranged missiles that denies the enemy the opportunity to effectively regroup and provides escalation dominance. This approach relies on a nation's best men being loyal and patriotic enough to lead units in very dangerous environments and also to push the limits of missile and missile manufacturing technology.

So far, the latter doctrine has done much better. Proponents of the former doctrine believe that LLMs will suddenly reach a tipping point where they go from an incompetent military liability to an asset. Personally, I do not foresee a future where chatbots trained on reddit data ever demonstrate an emergent understanding of how warfare works in practice.