Don Codeleone

im starting to think the 'distillation attack' claims are complete cope. turns out identifying whats distillation and whats simply the outcome of training on the same universal dataset (aka the internet) is pretty hard. i want to believe kimi k3 distilled fable 5 but how on earth could they have done that when the model was banned up until a few weeks ago?

Therapy has become incredibly dangerous for societal bonds. In a new study, 1/3 of parents believe their child won’t speak to them because of a therapist’s influence. This is how countries weaken and fall apart. Rome didn’t fall primarily because of barbarians at the gates. It fell because loyalty, duty, bonds, and the birth rate collapsed. When the Romans lost the traditional family, its societal structure that fostered mutual support fell. Therapists are creating a mental health crisis by actively encouraging patients to cut off their families. When communities are no longer connected, they become unrecognizable to their people.

Children are the future

There are literally hundreds of benchmarks for AI models, so why are there zero benchmarks for harnesses? I honestly think that’s way more important. You can seamlessly run other models inside Claude Code or Codex anyway.

At this point, most $20 subscriptions give you less usage than a free trial. So here’s my question: How are people in countries outside the US, Europe, or East Asia—where the average income is under $1,000—supposed to afford a $200 subscription? And what worries me even more is that it’s only getting more expensive.

Lets try Gemini 3.6 Flash

My computer tech teacher doesn't know what Claude Code is. This is why school is inferior to entrepreneurship. Lock in.

It would be epochally stupid for America to try and ban superior foreign AI. First, I suspect it's a violation of Freedom of Speech (you know, one of our most sacred values). But second, and more to the point, it would be dramatically financially harmful to the AI landscape. In the long run, no one gives a shit about frontier AI labs. "Commercialize the complement." The real goal is selling more chips. Data centers are the hard currency of the future (or, more specifically, GPUs and accelerators). And we have more. Doesn't matter whose model is running on it. The data centers are coming. AI models were *always* going to be commoditized.

I’ve said this many times before, but I’ll say it once more: The AI race is a marathon, not a sprint. Having the smartest model right now doesn't mean you'll be the first to reach AGI. Last year, Gemini was the top model and Claude was terrible. Real-time rankings will keep shifting, and companies with solid fundamentals and long-term stamina hold the real advantage. To me, that comes down to the vertical stack—and the ones holding the strongest stacks right now are xAI, Alibaba, and Google. On top of that, Chinese AI labs have a clear structural advantage in building these stacks due to policy support.