i see this sentiment a lot - what exactly is it referring to? Quotes by senior staff or actual ads?

None of this is a secret either - these employees openly talk about this on twitter.

Also the modal openAI employee at leadership level has been a maths nerd worrying about shrimp safety and the AI becoming skynet- not some rich dudebro.

I think you underestimate how difficult it is to get the models to do what you want them to do.

The obsequiousness is a thorn in OpenAI's side - not something they're trying to get out of their AI girlfriends

you can't port over the same stereotypes of 2010s and 20s era product managers or growth hackers at facebook to this group of largely maths, physics majors who were working at universities for much of the 2010s

That's the part that imparted the obsequiousness, and the AI labs and their employees largely hate it.

"Tech bros" is also a bit of a worry - all the top AI labs combined only employ maybe 10,000-20,000 people and only a portion work on this part of the models

would highly recommend vetting this further - as relayed in this thread it's a very flawed understanding.

The RLHF (reinforcement learning from human feedback) isn't done by tech bros, it's done en masse by contractors in poorer countries like Nigeria, The Phillipines etc.

except that they aren't the ones doing the RLHF that made it like this (that would be underpaid workers in Kenya, Uganda, the Philliphines etc)

The profitable stuff (increasing productivity for companies who will pay large amounts of money, or replacing workers) is much further behind.

It's best adoption & use is in coding, but a lot of that is becoming commoditised.

I don't think this is right, though.

The biggest problems caused by AI right now (kids cheating on homework, deepfakes, spam and scams) are not that profitable for AI companies.

wait if you block - other people can't see replies? That's wild

because that's roughly who has written stuff on the internet for the past ~30 years, and who writes non-fiction-y stuff, and news articles etc.

could the design be made by AI and then printed explaining the weird text?

australia locked down early and hard, with incredible compliance, and we were locked down for longer afterwards than the US was.

Australia dropped the ball on vaccines which is part of the story, but it wasn't some miraculous cure that let people out earlier

yeah - and we have far more state capacity and obedience.

Their training data is just everything ever written on the internet or otherwise.

They're secretive because of potential lawsuits, depending on how copyright law is applied.

so gorgeous man, congrats

This is basically wrong thankfully - the reason it's "hamfisted" is partly because there just doesn't seem to be good ways to influence the AI.

The training data is robustly centre-left, and it's very hard to change that afterwards