Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Computer, Respond to This Email

googleresearch.blogspot.com

Post reply on HN
Print view View original

171–180 of 202 posts

Re: Computer, Respond to This Email

#171

Post by kobayashi »

anemic wrote: post #94
dikaiosune wrote: post #4

While the technology is very cool, this pushes me that much closer to a "dumber" email service. It's always a conflict for me as I love shiny new things, but I'd rather we let AI loose on someone else's email, especially when the AI's revenue stream is advertising (a business predicated on knowing as much as possible about your audience).

advertising becomes scary when this service tries to suggest responses like "Yes, let's have a meeting tomorrow and why not enjoy a refreshing glass of ice-cold Dr.Pepper together!". But if not, I'm not that worried about the ad revenue aspect.

Please drink a verification can.

Re: Computer, Respond to This Email

#172

Post by _smaugh »

techdragon wrote: post #168
rylee wrote: post #127

Earlier quoted context omitted.

You can still do this (at least in Windows 7, the last time I checked.) Just look in the 'Mouse' section of your control panel, and select the one to make radiate circles around your mouse :)

I can confirm it still works in Windows 8/8.1

also in 10. Didn't know about this feature. Thanks

Re: Computer, Respond to This Email

#173

Post by Houshalter »

imh wrote: post #29

I appreciate the privacy standards they used (no humans reading your email to develop this), but am concerned that it's not enough. As I understand with language models, overfitting takes the form of returning a sequence of words seen in the training set. If this is overfitting in any part of the response space, this could happen. Out of a million emails, how many suggested responses are going to substantively resemb…

Overfitting requires "memorizing" the dataset, instead of generalizing it. I think that's very very unlikely. The neural network parameters can only store so many bits of information. But the dataset is millions of times bigger.

Re: Computer, Respond to This Email

#174

Post by Houshalter »

smtddr wrote: post #108
dikaiosune wrote: post #4

While the technology is very cool, this pushes me that much closer to a "dumber" email service. It's always a conflict for me as I love shiny new things, but I'd rather we let AI loose on someone else's email, especially when the AI's revenue stream is advertising (a business predicated on knowing as much as possible about your audience).

Just fyi, https://mail.google.com/mail/?ui=html ...still works! I bet there are some people on HN who never saw the original Gmail.

Wow that look so much nicer than the real gmail. And it loads 10x faster too.

Re: Computer, Respond to This Email

#176

Post by shrikrishna »

In a parodical reality with mistaken pressings of the "two buttons to reply" feature

> Obama: Hey Hillary, what do you think, should we bomb ?

> Hillary: Sure! Count me in!

> Hillary: Hey! Don't consider that last email! Damn auto reply...

That aside, as a fellow nerd, neural net FTW!

Re: Computer, Respond to This Email

#177

Post by gosub »

jes5199 wrote: post #175

Did they really write this whole article about machine-generated text but not actually include an example of the actual reply the system generated?

There are three example responses at the bottom of each picture, above the navigation buttons.

Re: Computer, Respond to This Email

#178

Post by lwf »

jes5199 wrote: post #175

Did they really write this whole article about machine-generated text but not actually include an example of the actual reply the system generated?

They did, in the three blue text areas below each message in the screenshot.

Re: Computer, Respond to This Email

#179

Post by Natanael_L »

blixt wrote: post #124

This kind of stuff is really cool. I imagine the future of Google being the same old search box, but instead of entering a search query, you engage in a conversation with Google so you can delve deeply into a narrow topic and get back tailored responses to your question (as opposed to opening 10 tabs of Stack Overflow links that are maybe related to your question). I worry a bit about the long-term risks of this kind…

Essentially integrating Wolfram Alpha and IBM Watson into Google?

Re: Computer, Respond to This Email

#180

Post by TeMPOraL »

microtonal wrote: post #170
TeMPOraL wrote: post #154

Earlier quoted context omitted.

So basically, because pretty much every useful data-and-automation-based thing - especially on the social, not individual level - can be considered an erosion of privacy, should we just stop progress? That's the sentiment I feel radiating from yours and others' complaints about privacy here.

should we just stop progress? I would love if companies would explore on-device processing more. While it might not work in this case, it should work fine for e.g. data extraction for flight tracking, hotel bookings, package tracking, etc.

Oh yes, so much this. The cloud business model thing leads to not just privacy-eroding solutions, but to implementations that are absolutely ridiculous from the engineering point of view, and exist only because the company thought it's a good way to extracting rent from the user. See, for instance, most of the work done around IoT, home automation and by hardware startups.
Post reply on HN