Tweets
2023-01-28T18:31:51+00:00 | 🔗
@marceldotsci Docker! We made the various conscious decision to move to docker after losing days to TF/PyTorch/PyTorch Geometric/CUDA/RDKit versioning
2023-01-26T16:33:20+00:00 | 🔗
@simonduerr The results of EUOS/SLAS were similarly interesting around 0.1 kappa score was reachable by normal methods and it seems like that you could get up to 0.2-0.3 by accounting for plate effects by inferring the experimental setup/id relationship
2023-01-23T14:40:07+00:00 | 🔗
@goodside For people who say that... how is prompt engineering different from providing context? writing a brief or a memo or a proposal? There's a reason why we have so many conventions for writing ~ documents. Sure we can work without them, but it can be vastly inefficient
2023-01-22T05:51:22+00:00 | 🔗
Very pleased with how this turned out! I’ve really been liking gouache. Probably still need to go over it a few more times, though. https://t.co/woI1WyVLun
2023-01-18T00:52:00+00:00 | 🔗
@svpino @fahadsamman IMO lukewarm take, I want OCR to recognize what I wrote in MY notebooks, not what YOU think I wrote… it should be about predicting what the writer intend not what random people think at first glance :)
2023-01-15T17:55:57+00:00 | 🔗
@RichardMCNgo Everyone overestimates what they can do in a day but underestimates what they can do in a month.
2023-01-14T23:56:33+00:00 | 🔗
@EMostaque @Midjiwan Polytopia is incredible! I’ve been just chasing high scores with Kickoo.
2023-01-10T17:40:49+00:00 | 🔗
@Sim_Startup_ @OpenAI Depends on how much of a stake they already have
2023-01-05T17:01:37+00:00 | 🔗
@andrewwhite01 @wpwalters Funnily enough in testing, I would say the most published DNN methods are out performed by Random Forests + XGBoost/LGBM etc. given proper choice of descriptors and ensembling.