honnibal Some under-appreciated reasons why CPU is strong for many NLP tasks:
1. Batch sizes often need to be small, especially when objective is non-stationary (e.g. for imitation learning)
2. Caching stuff is awesome for NLP, because Zipf's law. twitter.com/yoavgo/status/…
kathrinpassig "Wo möchte der Rollstuhl denn aussteigen?" Ich glaube, der Bus braucht eine Weiterbildung.
Wort des Tages: Mettsmoothie (h/t @kfl02)
evanmiltenburg #LaTeX tip: if you’re aligning text in TikZ (and sometimes in running text as well), \vphantom and \hphantom are your friend!
\hphantom{} is exactly the width of its text argument. (A phantom word!)
\vphantom{} is exactly the height of its text argument.
EnglishGibson ‘Sozial inadäquate Verhaltensweisen’
Two things to say about the above:
1. ‘inadäquat’ translates as inappropriate, not inadequate;
2. the main purpose of ‘-weisen’ is, arguably, to make ‘Verhalten’ plural.
grauhut Ich habe bei einem Kunden COBOL Code gefunden. Gut. Mainframe. Nichts ungewöhnliches.
Der letzte Kommentar ist von 1985.
Von meiner Mutter.
Herr_Rau "A Rick Brant Electronic Adventure" (1947). Zehn Jahre später wäre das wohl "atomic adventure", heute: "digital". pic.twitter.com/Kyj8iZUmMI
Research_Tim (9/11) On top of all that there is the big issue that science is inherently a team effort, yet most rewards are not. This is another example of how the system is not aligned with actual practice and the things we should be wanting to incentivize and reward. #APS18SF
Research_Tim (6/11) ... so many more scientists are doing good science than winning the esteem lottery. In addition, most rewards seem to be heavily focused not on good scientific methods but on winning the 'very interesting results' lottery. #APS18SF
Research_Tim (4/11) Is that necessarily bad? I tend to think so, at least in the case of academia. Esteem is a highly unequally distributed currency, inherently susceptible to unchecked inflation - its value is not bound to any physical commodity but primarily by our perception of it.#APS18SF
Research_Tim (2/11) What do I mean? It seems that in academia there is only one currency: esteem. Essentially all incentives play into this, all rewards are proxies of it - publications, citations, grants, awards, keynotes, you name it. #APS18SF
jason_a_w it's not news i guess but these tech guys clearly think that they are better placed than anyone else to solve any given problem
jason_a_w The ides that white supremacy, white nationalism, and white separatism are clear and distinct ideological positions is one problem. Also their examples as to how people might express that suggest that they have never logged into their own website.
jason_a_w goodness me, check out Facebook's training advice to moderators on dealing with racists motherboard.vice.com/en_us/article/… pic.twitter.com/XjY5Lg9AG3
elnathan_john Dear homophobic Nigerians,
Hating another human for who they are and for something as integral to their being as their gender or sexuality is not "seeing things differently" or "having a different opinion". It is hate. It is violence. And it has no place in civilized society.
HummelFamilie "Hallo, mein Schatz."
"Hallo, mein Schatz!"
Kuss.
K2: "Heißt Ihr beide gleich?"
skeptikantin Entweder hat „rabiat“ im Schweizerdeutschen eine gänzliche andere Bedeutung oder ich bin hier sehr, sehr richtig.
watson.ch/Schweiz/Ostsch…