[virgilp]

Yes. Romania. What you think software didn't run in Romania?

What did you use, Dragon Speech? Maybe it worked better for you? I can assure you for non-native speakers it was pure crap. Also, what was the usage in western world? I know they had _some_ sales, but I don't remember it exactly taking off.

Lastly - you're obviously willfully ignoring progress, like translation for instance works so much better with RNNs and attention. The basics of RNN were indeed there (LSTM was just published - though e.g. GRU is more recent), but attention was not. And attention is critical. That's just one of the things that's not "basically the same" and "worked just as well 20 years ago".

[scottlocklin]

Yeah, we had a guy at LBNL who was a big OS2 fan. Worked fine for me, and I do have a fairly heavy coastal accent. Mind you it worked in a big, loud industrial facility.

I thought the previous KNN based approaches worked better than the RNN based ones. Which is more or less my point here. The main piece of "progress" we got was the result of a giant corporation having a large corpus of data. The actual technology hasn't improved or changed a whole hell of a lot.

I'm eventually going to write a history of technological progress documenting this. Progress is very obviously slowing down other than the improvements in lithography which are now fading out. Throw away anything with a screen in your household and life is basically identical to what it was 20, even 40 years ago. That wasn't true within my lifetime: there were huge strides made in technology in decades gone by -in my lifetime even. We're at the point where a lot of "progress" is just changes in fashion; KNN versus neural approaches. People buying the latest iphone (which is really no more capable than the early iphones); wondering at the facial recognition IR thingee. Oh sure, DL can do some tricks that were not possible before, but their actual real world impact is marginal, and sometimes, it's not even an improvement over other techniques.

[virgilp]

Knn works better with little data, rnn with lots of data. But we do have lots of data - that, too, is part of the progress. Also progress is non-linear; and the real world impact is huge, though we may not always like it (social media & content algorithms have literally changed the world, influenced elections and sparked literal revolutions). Driverless trucks would be a huge shift, and it’s not unconcievable that we’ll have them. RPA is threatening a lot of repetitive white collar jobs; sure it’s boring technology, but that’s exactly what makes it disruptive in my mind. Image understanding _did_ get a lot bettter (face recognition along with it) and we still haven’t fully processed the implications of that.

You have grounds to be skeptical about the future and to believe it is dark; but IMO you don’t have grounds to claim that there was no progress.

[scottlocklin]

I don't claim there is NO progress; just that it's extremely weak compared to the huge triumphs of, say 1820 -> 1970.

re: face recognition: Viola Jones is almost 20 years old (we've been talking about 20 years periods). Eigenfaces is over 30 years old.

I don't believe in Driverless trucks, and it is emblematic of current year whenever I talk about technological progress, people bring up something that only marketing departments claim actually exists in the corporeal world.

FWIIW by "KNN" I don't mean the O(n^2) classic; the thing Google was using before was a bit more sophisticated than that, and it worked just fine. For some languages it worked better than the "advanced" RNN version of things.

[virgilp]

Now I'm starting to doubt your knowledge.

How exactly do you do face recognition with Viola Jones? Nevermind an accurate-enough version to be used for biometric identification... just one that can distinguish you in your photo collection.