prlambert

I was the PM on this feature (and these are my views not Googles, as per normal). The truth is actually much more banal. Most people are just really positive over email.

The model is trained to offer reply suggestions that have the highest chance of being accepted, from a large whitelist of the most common short replies. The whitelist contains many negative options. We're optimizing for click through rate. That's it. There's no editorial judgement, definitely not "‘no’ struck someone as too negative and they had to take it out."

We actually experimented with intentionally inserting more negative options to increase diversity. Doing this reliably causes a hit to our metrics.

Discovering this made me pretty happy about the world. Most people are generally pretty friendly to one another (at least over email!).

(also @reaperducer) Something I didn't fully appreciate until I got here was: 1. Work (aka enterprise) sends much more email than consumer, and 2. Outlook dominates enterprise. Their market share is pretty astounding.

I assure you there is a lot of email coming from Outlook/Exchange.

I'm genuinely sorry to hear that. Our analysts are looking into the issue described in the article, hopefully we find something we can fix that will resolve this.

I'm one of the PMs for Gmail and hang around HN quite a bit. This is my personal take, not an official reply.

It's simply not true we have no incentive to fix this. Here are a few:

Firstly, Gmail's success is entirely predicated on the health of the global email ecosystem. Gmail does not, inherently, have any network effects (unlike FB, messengers, any other comms tool of Gmail's scale). Email itself, of course has a huge network effect, and that is because you can email anyone in the world, regardless of what email system they use. It's because email is open. If we lose an open, healthy ecosystem with many providers, we'll destroy the base we stand on.

Secondly, we care deeply about having positive relationships with developers and all our users. I can tell you it definitely makes me sad to see articles like this. There are going to be false positives, we will make mistakes, but we certainly care a lot about fixing issues like this when we hear about them.

I agree Postmaster tools has been underinvested in and we could do much better there.

Fair enough! thanks for keeping me honest. Unfortunately edit window has passed so can't fix as suggested in the guidelines:future readers please ignore the question and start at sentence 2.

Did you read the post? He specifically calls out biographies as the exception to the 'no journalists' rule and describes why journalists are well suited to write bios.

In other words there's no reason to believe the author would miss any of the books you mentioned by his "journalist disdain".

"NYC has a lot of developers, more than the valley" - do you have a source for that? Seems highly unlikely. My quick googling suggests silicon valley has 2x-5x more engineers than NYC.

Thanks for all the feedback! I can address of few of these.

1. Threads should be sorted by date of most recent email, regardless of importance. This sounds like a real bug and I'll dig into it.

2. Yes, great feedback. You could archive it, does that work?

3. What's the use case? Why do you want to keep spam? You could forward it to yourself and it should work (a bit hacky, i admit)

The last 3 are all not going to happen for a number of hard reasons, unfortunately.

Have you tried multiple inboxes in gmail? You can create custom sections for specific labels and then sort your incoming mail into them (using keyboard shortcuts this can be really quick). For example you can have inbox sections: Starred P1 P2 P3 Inbox (new/untriaged)

You can also use 'super stars' to accomplish multiple priority states within starring.

Here's an example of a real power user: https://medium.com/@gregsramblings/how-i-tamed-gmail-at-work...

You can disable Smart Reply from the mobile apps, the setting isn't there on Web yet. Honestly we just haven't prioritized it yet.

Here's some info on how it works if it helps address your concern: We've transformed the language generation problem into a labelling problem. From a systems point of view it's not any different from spam filtering, or tabbed inbox, and we use the exact same features. Smart Reply uses a whitelist of around 30k independently generated & sanitized phrases. You could think of each whitelist entry as a label. So when an email comes in on the delivery pipeline we just label things, like we've always done. We label (spam/not-spam) for spam filtering, we label (promo/social/update/forum) for tabbed inbox, and we label (Thanks!/Sounds good!/etc/etc) for smart reply.

Could you be more specific? Which features would you like to see?

(I'm the PM on Gmail for intelligence and filtering)

Based on Dan's research we built a new gmail feature: high priority notifications. It only sends you a push notification for really important messages, cuts out over 90% of Gmail notifications for most users. https://www.google.com/amp/s/mashable.com/2018/06/15/gmail-a...

Importance is defined by a model trained on reply rate and a few other engagement metrics.

Edit: High priority notifications is now available for both Android and iOS, contrary to what the link I included says

I'm curious as to how that breaks down? Which kind of workflows are best in email, which in Slack?

Sorry to hear that, I might be able to help. I work on Gmail. DM me the details on twitter @prlambert (or if you gain email access prl@google.com).

Rolling out over the course of the day. We'll be at 100% in a couple hours. Thanks for your patience. :)

It's a great question and one of my early concerns as well.

For most people email is a utility. The diversity of language in the middle chunk of the distribution isn't very high today and the biggest complaint most people have about email is how much time it takes. Most of our users aren't writing poetry, they are doing every day business transactions and we can help them be much more productive. Think how many times you've written "hope to hear from you soon", or some equivalent.

For the tails of the distribution Smart Compose is not helpful. We address this primarily with a triggering model, we aim to only show suggestions when we're quite confident that you're in the 'just getting things done' mode.

And if you're the type of person who always has a lot of personality in your emails, the feature probably isn't a great fit for you. Today it's opt-in, and there will always be a setting even if we turn it on by default.

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.

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.

This is correct.

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

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.

The language itself isn't 'more cool' or better, but the risk profile of adopting it has dropped dramatically. That makes it much more interesting.

I was the PM on this, happy to answer any easy questions.

They would if they could and it was clean/legal. But there are very few purchasable databases that make a dent at Twitter or LinkedIn scale.

This line of thinking was behind the original motivation for Digits (part of the Fabric SDK). Get developers to collect Address Books for us, in return they get a free SMS-based verification service.

I was the PM for this at Twitter. You pretty much nailed it. The straight Address Book algorithms are Reverse Address Book (your #1) and Strong-ties prediction (your #2 and #3).

LinkedIn for sure also uses common groups (schools, workplaces, interests, etc), and geolocation in their recos. There's also sneakier signals like app graph and browsing history, though in my experience those were more useful for content recos than people recos.

Interesting to see this get traction here today, this article is for the web version which has been around for awhile.

But we did just (in March) launch this for Android: https://blog.google/products/gmail/send-and-request-money-in... & https://techcrunch.com/2017/03/14/you-can-now-send-and-reque...

PM @ Google, before that PM @ Twitter for 2 years. Before that CEO/Founder of a SaaS startup.

Did a BA in Anthropology & English. Learned to write, think clearly, and communicate effectively.

Moved to China, founded a company which failed but gave me the entrepreneurial bug and the realization that I didn't know nearly as much CS as I thought I did.

Went back to school, did a Bachelors of Computer Science w/ Math minor. Learned to think systematically & rigorously, how computers work, a bit of ML/algorithms/etc, and how to code.

Applied to Google APM program. Thought it would be an ideal job before founding a company. Didn't get in.

Founded a company with my best friend, who had a Masters in Computer Engineering and had worked in industry. Spent the first year coding everyday, all our first employees were engineers. Learned to write real software and work with an engineering team.

Transitioned into the CEO role in practice. This happened because my co-founder and I had very different motivations. He wanted to build great systems (and not have to go to meetings / talk to customers). I just wanted to do whatever was needed to move the business forward, I was equally happy coding or talking to investors / taking meetings. Learned (through multiple failings) how to think about markets, interview customers, how to motivate and lead, build realistic roadmaps, raise money, do sales.

After 5 years, sold my company as a 1st base hit exit. I had no problem getting mid-career PM interviews, despite having never been 'officially'a PM. Focused on big scale consumer because I wanted maximize my leverage/impact.

Junior positions? See todd_sherman's response (Hi Todd!).

Skills – well covered elsewhere. The short version is "Get the right things built & shipped." To do that, you need to know (1) what the right things are (user problems, strategy, markets, corporate politics, etc) and (2) get them built & shipped (working with eng & design, leadership, resourcing, more politics). It's harder than it sounds, there are many many more wrong things than right things (and many wrong things can sound pretty right).

I'm sure I'm biased, but I think founding a company is an ideal pre-PM curriculum. When big tech co's acquire companies, the CEOs almost always become PMs, for good reason. This knowledge changes the risk equation as a founder, being a PM at a great company is a pretty awesome backup plan.

Yes this is normal. I work at Twitter and new fathers taking multi-month paternity is very common. Lots of friends at Google and believe it's the same there.

Yes.

Once someone is on your cap table it's very very hard to get them off. If it's someone who isn't reliable, doesn't align with you philosophically, or is unethical.... There's a world of hurt waiting. You spend the money once, you deal with the consequences for the life of the company. Little things include having to chase down signatures years after they've "checked out" so you can run your business. Bigger things include blocking an exit for petty reasons, or scaring away future, higher quality, investors, because good investors don't want to be in deals with bad ones.

Happy to answer any questions, I'm around.