@DrPhiltill@quasistable Take as an example: the USPSTF's decision to recommend against PSA screening in older men. This is exactly the sort of thing people are complaining about on Twitter. It's not like they did their math incorrectly. They just made bad assumptions in their model.
@quasistable I doubt it's primarily a population dataset play. It's that they can solve the inverse problem now, when that wasn't possible a few years ago. First, there's way more compute. Second, diffusion models help solve the inverse problem, see for example https://t.co/scqoorvCAM
@thingsshldwrk@Robotbeat The USPSTF and insurance companies do these analyses regularly. The problem is really the assumptions baked into the analyses (e.g. bad follow-ups), and that at the end of the day the public pays for it (either through higher insurance premiums or taxes for Medicare).
@roddux@bl4sty Hey guys, I recently made a XML parser that is a drop-in replacement for libxml2 (my understanding is it's unmaintained now). It is actually faster and better in a number of ways too. You can read more and if interested, get in touch at
https://t.co/YOA6q5P7Dw
@RoseSilicon@NathanBLawrence I recently made a best-in-class XML parser that is a drop-in replacement for libxml2 (my understanding is it's unmaintained now). You can read more and get in touch at
https://t.co/YOA6q5P7Dw
@RebalAid@vildavedo@Helenrug@ProtonSupport I recently made a best-in-class XML parser that is a drop-in replacement for libxml2 (my understanding is it's unmaintained now). You can read more and get in touch at
https://t.co/YOA6q5P7Dw
@mimrocker@Steve_Yegge Google has a fairly good solution for large companies. They just interview people, but keep track of interviewer data. In theory, they're trying to figure out interviewer bias (who is hard, who is biased, etc) and then adjust for that. Google isn't perfect but it could be worse.
@cornu__copia I'm guessing most of the people there weren't actually property owners? Like in a favela where they don't have a deed, but just put up a building anyway.
@SwipeWright It feels similar to just a "Note" type response, or PubPeer, but I think it could be more useful by being organized the way you're proposing.
@aphysicist Anything that can be figured out in silico will be figured out first. Then with robotic automation, we'll start figuring out real world stuff too. CROs already use automation, there's already stuff like Emerald Cloud Labs, it's just a matter of making progress. It's the future.
@willmcgugan@AnthropicAI A more likely explanation is that Anthropic is experiencing hyper-growth, and thus has more important things to work on. Their APIs have had intermittent outages all week! They must be scrambling internally all over the place on high value work.