@AlexisCompBio @facontidavide Does this *landmark* study control for quantity of comments in general? My hypothesis is code with swear words just correlates with code with good, human readable comments and variable names, which then correlates with good code quality.
Calling #Robotics#Sensors#Autonomous#Offshore organisations. We are launching a new £1.5m competition in collaboration with DEFRA to improve assessment and evaluation of the UK's marine natural capital assets on 31st August. Registration link below.
https://t.co/D5MrkbtyVZ
Spending part of your summer holiday decorating? It's great to be supporting young #startups like @Hausbots develop #robots that can take a dull and dangerous job like painting at height and give you a robot to do it instead. https://t.co/NHNzbocSg3
(1/N) Yesterday @auterion published a report sharing survey results about US consumer perceptions on drone delivery. Short and infograph-laden so feel free to dig in. I'll share a few interesting tidbits in this thread. https://t.co/SBvetWMctJ
I’m thrilled and honoured to be appointed the first Chair of @ARIA_research - and to have the brilliant @ilangur as our founding CEO. A quick thread on why I think this is so important…
https://t.co/Tt0Qv8hH1X
@amypeace “A climate scientist looking worried at the weather forecast” - yup they are all men. Quite a bit of gender bias throughout, seen consistent male-only imagery for a variety of stem roles. OpenAI are pretty open about this and make no illusions their model is perfect.
https://t.co/uZjZl3PyEC I got access to #dalle2 this week. Fascinated by all its capabilities but equally interested by where it falls over. If you try “a dog with six legs” or “a dog with six legs, not four” it consistently spits out just ordinary dogs.
@amypeace So. “War” is against content policy, but take that out and you get these. Definitely captures the “thou shalt recycle” essence, but DALLE2 is bad at text, even to the extent of not knowing what alphabet to use.
@amypeace I signed up for the access as soon as they launched, used personal not work address so definitely not special! only just got access so it was a few months wait. Style transfer or changing textures like the example below works really well.
@the_accidental Interestingly it doesn’t generate images of dogs who have obviously lost a leg, instead four legged dog with a fully or partially occluded leg.
@the_accidental It���s really interesting because of the huge feats of ‘imagination’ the model is capable of achieving, but some concepts must be too ingrained to ‘un-do’ - especially if there are no examples in training data. I reckon three legged dog is fine.