Gilbert

“Counter factual” is so last year, this year is all about “Uncertainty” #ai #ml #papers

It’s good to hear tf 2.0 is out . But “driven by community”? It’s more like in tf 1.0 the last thing we cared about was community now we’re losing so much traction we have no choice. #contribisajoke #longlivekeras #sonnet

Never figured out how there are ML folks who claim they do everything (vision, language, time series, speech, etc..). Guess all the crazy know-all job descriptions are real. I prefer the type that actually tell you they specialize in one domain/ problem. #mlhype #aihype #ai

How do we know that we’re not on the path to becoming mars? (Due to global warming). Every earth like star/ dwarf etc is a once upon a time variant of earth that also suffered from climate change due to similar reasons. @elonmusk #simulation @neiltyson

Simpler things than above fool humans. Super human level trademark is meaningful only in context it is a match to us humans. But doesn’t say anything about intelligence as well.

Interaction between them like a routing model like capsules this would still be narrow ai. Even if it’ll have context tracking and what not. This is still narrow ai. Not agi, more like Siri Alexa google assistant

it even more, it wouldn’t change to you up to some reduction of the space. An intelligent model is one that can do this compression efficiently. This is narrow ai. Beyond human level. If tomorrow we’d have multiple models such as that across multiple domains and tasks with

@AutoMLFreiburg @jeremyphoward @JeffDean

Automl can help find a better lower bound on capacity than us, or more generalizable model. Which is huge. The question is: can we interpret the finding? If not, does it matter? And can it find a novel symbolic architecture that wasn’t designed by humans for it to use.