Not to mention beaten Go, pro-level Starcraft, etc
-
-
-
Loling because there’s a bunch of academics agreeing with me and a bunch disagreeing. But my everyday experience doesn’t deceive : my phone scans my face to let me in and I can talk to it reliably
Prikaži ovu nit -
Another indication CV is solved: as OpenAI and DeepMind grapple with real time games, their agents directly interface with API observations instead of reading the gamescreen: they know it can be done, so they’d rather save the compute time
Prikaži ovu nit -
Presented without comment https://twitter.com/amartincastro/status/1223407933205037057?s=21 …https://twitter.com/amartincastro/status/1223407933205037057 …
0:10Prikaži ovu nit -
Presented Without Comment https://twitter.com/verge/status/1223737308106625026?s=21 …https://twitter.com/verge/status/1223737308106625026 …
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
-
-
Computer vision isn’t solved and neither is speech recognition. Nothing has been solved completely yet. Great progress tho!
-
>99% on image net is pretty good!
- Još 17 drugih odgovora
Novi razgovor -
-
-
What do you think the next breakthrough will be?
-
If I knew this I’d be working on it myself :)
- Još 1 odgovor
Novi razgovor -
-
-
hey, so the state of the field is that: - pretty much all CV systems still will never spot a cow on a beach (not in the training set), - english SR on accents outside of WASP / British is wow bad, - translation only works well for the 3-5 language pairs that have major coverage.
-
Correct me if I’m wrong, but most of these seem like a train set augmentation issue? the limitation is with the data economics. when there is sufficient demand for some currently out of distribution classifier, someone will collect the data
- Još 2 druga odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

reinforcement learning research