For example, why does Google handle this differently for consumer and enterprise accounts? Well it's Google so the answer could always just be "they are disorganized" but there's a good chance that in both cases, it was the pragmatic choice given the slightly different priorities of these types of customers.

Not detracting from the article, I think it's a fun way to shake your brain into the entirely appropriate space of "rapid change is possible"!

Of course in practice I think the author is actually correct - LLM's struggle more than humans with sophisticated formal constraints and less than humans with remembering to write a bunch of boilerplate. But I think it's a pretty counterintuitive result and I'd love to have seen more discussion of it.

People outside the US prefer telegram because they assume that Signal is probably compromised, or at least highly vulnerable to compromise, by US intelligence - they trust Pavel Durov's history of expropriation and arrest more than they trust some nerds who claim that our product is secure.

I think its obvious that making the game should be legal, and also obvious that platforms like Steam should ban it. Payment processors are a weird middle ground, I'll leave it to smarter people than me to figure out the ethics there.

Anyway whether or not this tool is advisable its definitely cool, nice work!

As far as goalpost-moving goes, it's wild to me that nobody is talking about the turing test these days.

OpenAI has also invested heavily in human annotation and RLHF. If all DeepSeek wanted was a proxy for scraped training data, they'd probably just scrape it themselves. Using existing RLHF'd models as replacement for expensive humans in the training loop is the real game changer for anyone trying to replicate these results.