How should research communities act? While my feelings about the ethics of surveillance tech are strong, there's a need to ensure that harms are not just hand waved away, but are a core part of evaluation. This wouldn't fix problems like racial bias, but it would at least throw it into stark relief.

There's definitely a fair number of AI researchers here on Bluesky - while there are some blinkered boosters, there's also a strong community of people here with a more balanced expertise and interest. If you're interested, feel free to reach out.

How's that even going to work with the 2x DP cap, given +1 yr on responses? Seems as well thought through as a paper raincoat.

ECR/MCR precarity is already ridiculous, but this is only going to make it worse. Genuinely have to wonder if it'd be easier for the ARC if we all just left research.

How's that even going to work with the 2x DP cap, given +1 yr on responses? Seems as well thought through as a paper raincoat.

ECR/MCR precarity is already ridiculous, but this is only going to make it worse. Genuinely have to wonder if it'd be easier for the ARC if we all just left research.

That's the idea behind openDP though - to provide a product that stops the need for companies to roll their own closed source. It's the same as crypto - roll your own, and you're guaranteed to create vulnerabilities. As to speed of adoption? Hopefully maturity fixes this. But it exists.

OpenDP to me holds some hope - even if there are flaws now (no idea if there are, I'm out of date with my DP implementation knowledge at the moment), but more due to their philosophy of implementation, focusing on probability and verifiability rather than SOTA.

More broadly, corporatised education providers centralize power, with parents feeling obliged to deal with companies to provide educational services that their kids otherwise aren't able to access through school.

1000% - or just spend the time you were thinking about that changing your passwords. Literally anything seems like better risk management.

First year grad students are being added to review pools. Hell, Neurips added high school students to the review pool. So it may not be outsourced work, but rather the broader failures of peer review.

From a market competition perspective, is stopping training even going to be viable prior to the existence of a monopoly, or monopoly like players? If a LLM never can reach the point of being feature complete, in my eyes there's always going to be a competitive incentive to try new things training.

Sums it up.

These practices are still happening - a non-Aus Uni sent recruiters out to Australia recently, and all interviews were in a hotel suite.