Right now, AI detection bots misidentify English language text from neurodivergent people and people using English as a second language as AI. I suspect it's much worse in other languages -- too many Anglophones pretend that English is the only language worth supporting.

It's not about general competence or pride. Writing safe C++ is a human vigilance task, and humans suck at vigilance tasks. If you are practiced enough with C++, your brain can automate a lot of that work. If you're an excellent programmer with Java and C# and Python and JS and Erlang, and then you start writing C++, your brain can't automate that yet, and you either go very slowly and bother experts a lot, or go as fast as your manager wants you to and break things.

Generic popular videos, with "potentially sensitive" stuff filtered out and several slots reserved for hand-picked channels (usually MrBeast and SSSniperWolf in the US).

Yeah, even golf courses use more water per day than this data center uses per year. Mind, AI is an extravagance and should be penalized as long as it doesn't have a socially useful use case, but that's for a host of reasons beyond water use. Golf courses and growing alfalfa in a desert should be much higher on the chopping block for water use.

Extinction is also part of evolution. Humans have been demonstrating this by being the Holocene Extinction Event. We can choose to limit the event, but then the shareholders would lose value.

This is an improvement. Every step you take to make it more difficult means fewer people do it. Forcing people to go through hoops means they're more likely to run across stories about how inaccurate LLM output is.

Currently, they censor people for sexual content. By refusing to censor Nazis, they are saying they support Nazis more than people who want to post nudes. They prefer genocide over sex.

Communists want to guarantee access to basic resources. Nazis want to kill millions or billions of people.

You: I see no difference.

This is a "water is wet" article. Something that's widely known but needed an actual study to prove, and now other scientists can properly cite it.

Similarly, meta-analyses and introductory summaries tend to be very well cited, even though they're just telling everyone what they already knew.

Mastodon isn't funded by VCs. Eugen Rochko explicitly rejected VC funding.

They want globalization when it's USA's corporations taking over businesses in second and third world countries. It's imperialism. Always has been.

You can tell that China is a dictatorship of the working class because the police bust unions, right?

I wonder if it would be productive to analyze China in terms of mercantilism.

A large base model (roughly terabyte scale, I think). The file you receive is a transformer on top of that model plus some input data. It's a curiosity, not something we'll see in production in the next ten years.

You'd distribute the base model (100GB to 5TB) first since it's useful for a lot of inputs. Then you'd download files that have the transformer plus the compressed data.

The texts they're testing with are 1GB sections of English language Wikipedia dumps, and they're getting poor compression until they use very large transformers, like 100MB.

They also have some weird comparisons for partially decoded data and I don't get the point of it. Like, they show intermediate output of gzip that's basically text with the Burroughs-Wheeler transformation applied and point out that that's not very comprehensible compared to some "equivalent" from their compression system, and it's like, no shit. So this is pretty useless.

We're in another giant COVID wave and our employers want to force us into poorly ventilated boxes cheek by jowl with 30 other people, few of whom wear masks.

I work better from the office. I would commute by bike, and that's better for my health. I benefit from the automatic body doubling and having a focused zone for doing work.

But I won't benefit from getting COVID twice a year, so I refuse to go to the office.