Live data from Hacker News

Write Emails Faster with Smart Compose in Gmail

blog.google

101–110 of 200 posts

Re: Write Emails Faster with Smart Compose in Gmail

#101
post #43

Earlier quoted context omitted.

Two big problems. One, AI mediation of language sees issues like bias amplification (AI learns our biases then suggests them back to us). Google itself wrote about this recently [1]. Two, spell check has a homogenizing effect on spelling [2]. If you extend this to the level of phrases, that will homogenize not just spelling but the very way we talk. [1] https://developers.googleblog.com/2018/04/text-embedding-mod...…

To add a third, I have a problem with my personal emails being used as a corpus of text to train these models. Maybe they're using only my emails to train a model for me, and turning that feature off disables them doing so. Maybe they're using everyone's email to train a generic model. Of course, they've been (machine) reading your emails to provide ads since forever, so in that sense maybe this isn't really that new…

We're training Smart Compose from highly anonymized global corpus of consumer email.

We've stopped using emails for Ad targeting entirely, and the enforcement is quite strict, as reported in June: https://www.nytimes.com/2017/06/23/technology/gmail-ads.html

Re: Write Emails Faster with Smart Compose in Gmail

#102
post #74

Earlier quoted context omitted.

To add a third, I have a problem with my personal emails being used as a corpus of text to train these models. Maybe they're using only my emails to train a model for me, and turning that feature off disables them doing so. Maybe they're using everyone's email to train a generic model. Of course, they've been (machine) reading your emails to provide ads since forever, so in that sense maybe this isn't really that new…

Yup, I forgot where I read this, but I believe all Gmails are eligible for the model. Again, forgot where I saw this, but Google takes great precaution to protect your data from human inspection, but being used in the corpus is fair game.

This is correct.

Re: Write Emails Faster with Smart Compose in Gmail

#103
post #74

Earlier quoted context omitted.

Yup, I forgot where I read this, but I believe all Gmails are eligible for the model. Again, forgot where I saw this, but Google takes great precaution to protect your data from human inspection, but being used in the corpus is fair game.

I think if you are a (paid?) enterprise user, they do not do this. On the other hand, sussing out what is and is not fair game is quite a lot easier with a service such as Fastmail.

Paid domains can choose if they want us to use their data for training. Many do since it improves the quality of features they receive.

Re: Write Emails Faster with Smart Compose in Gmail

#104

I don't think a native English speaker would benefit much from this feature. But as someone for whom English is a second language, I am sure that I would have appreciated this when I was less fluent in English. I am now trying to learn French, and I find myself searching to the most trivial sentences in linguee.fr. Learning grammar and vocabulary is one thing; knowing the idiomatic way to express one's message is an…

This has been one of the most consistent themes since the first prototypes we tried. It is particularly helpful for non-native speakers, but the large majority of native speakers find it helpful too.

Re: Write Emails Faster with Smart Compose in Gmail

#106

I'm the PM for this and the author of the article. Happy to answer any questions, with the exception that I can't comment on future roadmap.

Hi, thanks for offering to answer questions. Did you consider the language homogenisation question and if so, how did you answer it?

Re: Write Emails Faster with Smart Compose in Gmail

#108

Earlier quoted context omitted.

If you write a word your spellchecker doesn't like, and you accede to the spellchecker, then it's not actually a word. Not to you. An utterance is only a word if you believe it in it and are confident enough to defend it against the opinion of a dumb machine. So this alleged culling of the lexicon by the spellchecker is largely a shakeout of non-words. Firefox's edit widget is underlining "spellchecker" in red, but I…

wait, so according to that, old words that are still written in old manuscripts/stones down which but we don't remember the meaning of, simply because they weren't used for too long, would not be words? I tend to disagree...

They are effectively like words in a foreign language.
Post reply on HN