another great regular online talk series! they're talking about GPT-3 now
@realSharonZhou: sees opportunities in medicine for with "democratization" of design of e.g. web interfaces.
this could be key for healthcare providers who have clinical expertise and know what patients need but don't have web design skills.
this could be key for healthcare providers who have clinical expertise and know what patients need but don't have web design skills.
@DrHughHarvey sees this as a step towards the holy grail of ML in radiology: a model that takes in an image and returns a full radiology report.
jump from GPT2 to GPT3 was just size. what might trillion-parameter models bring in other domains?
jump from GPT2 to GPT3 was just size. what might trillion-parameter models bring in other domains?
@DrLukeOR: COVID chest x-ray models benefit from pre-training due to large, public datasets with similar visual features. we need similar datasets for other diseases/modalities.
agreed!
agreed!
there's also the problem of training such large models. out-of-budget for typical academic units, even individual hospitals.
it's expensive, and @DrHughHarvey wants the NHS to get on it. sounds like a good idea to me!
it's expensive, and @DrHughHarvey wants the NHS to get on it. sounds like a good idea to me!
@judywawira: ML experts need to learn, from clinicians, how to better handle "messy" data like EMR.
well taken! not everything is as clean as MNIST (which itself isn't as clean as much of stat learning theory!) and ML folks need to get input from domain experts here
well taken! not everything is as clean as MNIST (which itself isn't as clean as much of stat learning theory!) and ML folks need to get input from domain experts here
@jeanne_shen: visions of a single AI-powered "cockpit of the future" for medicine need to be tempered, since different (sub-)specialties have different needs
to unify into one platform, "step 1 for a useful cockpit" is to _automatically_ "specialize to different sub-specialties"
to unify into one platform, "step 1 for a useful cockpit" is to _automatically_ "specialize to different sub-specialties"
sudden left turn: in the middle of @StanfordAIMI's AI Happy Hour, Sandra Lawrence of @TheSweetFarm introduces us to some rescued animals from factory farms or failed backyard farms, including "Feisty Fiona" the turkey.
i cannot emphasize enough how wholesome this content is
i cannot emphasize enough how wholesome this content is
