Massive flammability of the foam. If it starts in one place, and it's ordinary, non-flame-retardant foam, the whole ceiling will be burning aggressively within seconds. (Flammable claddings have caused other issues, like the Grenfell tower fire, but bar ceiling + small flame specifically happens over and over again)
It happened in Thailand at the Rong Beer Na Lat Phrao bar this summer, set off by a failing AC unit, ~30 dead ~70 injured. And in Switzerland at the Crans-Montana bar this winter, set off by a sparkler lifted too high up, 41 dead 115 injured.
If it was just this year, maybe it's a coincidence, but it was huge news back in 2003, when it was one of the first disasters widely filmed. That was in America at The Station nightclub, set off by band pyrotechnics, 100 people dead 230 injured.
As long as it's not across the whole ceiling of a bar.
It's a bit off topic, but damn. If you got a nickle for every time a double digit number of people were killed by that exact scenario, you'd have more than one nickle, which is ridiculous.
One thing that can happen: If your current friends demand something from you other than hanging out, kindness and maybe occasional favours, they're not really your friends.
I mean, maybe you still want to be in whatever tribe, but be aware that a cult or gang or industry network is a different thing.
That's true, if you had no GUI for an extended period, it'd be worth learning something that could do that. Although, other people are saying there's a way in nano.
If you're using a GUI, and working in an emulator, who cares if it's not "proper"?
There's lots of history you can look at to see how this works. The same thing does get invented more than once, but sometimes there's different technologies that could each become dominant, and then there's a real first mover advantage.
In Russia, they don't really have highways. Everything is built around rail lines. In the Inca empire, they had highways, but highways with stairs because there was no widespread wheel usage (which itself might be down to limited available wagon pullers and rough terrain).
There's totally open source models out there if you want to check what I'm saying, and lots of information on the training process, too. In mathematical terms, it's not complicated process.
People have tried to figure out what's going on inside, of course, and failed. Last I checked, the big advancement in interpretability was that someone figured out a tiny net that just adds two numbers.
So, what am "I" hoping for, I guess? More money, more possibility for advancement? What other jobs could I get?
Not to be a dick, but the median personal income for a full-time employee in the US is ~$55,000, and half are definitionally below that. If you live in San Fransisco and are working towards your PhD in data science, I guess sure, vomiting face, but that detail wasn't in there.
At the moment, whatever I’ve encountered seems to be acting as a data processor, although maybe they’re cooking something else we haven’t seen.
You've been misled, then. LLMs start totally blank, and nobody knows how they work later on. It's not like a normal program that follows rules somebody coded.
Even with memory, what would the AI want, besides what’s programmed and prompted?
To get the next token right. That's what gives the net a "pleasure" signal, implemented through backpropagation. Obviously, it grows the ability to handle subgoals of various kinds, but that's the most primal impulse.
Anything sticky and viscous would count, probably, so yes.
To be slime, you need a certain elasticity or gelatinousness as well. Honey isn't really slime.
Source: I made it the fuck up, but also, it seems right.