Show HN: Say Less – AI summarization tool in the Gmail compose window | Hacker News Reader

Show HN: Say Less – AI summarization tool in the Gmail compose window

sayless.email

135 pointsyoavz57 comments

I don't know if it's just me, but I very much prefer easy to skim emails to more dense ones.

At one point I got weekly project updates from a team I was working with, and one guy wrote dense, short, emails where I would have to read every sentence carefully to get a hang on what was going on. Another guy would write longer, fluffier emails but with bullet points and paragraphs in the same order:

> Hi everyone!

> {fluff}

> {general status}

> - Bullet points of things done that week

> {comment about things done}

> - Bullet points of things to do next week

> {comment about potential problems etc}

> {fluff}

> {annoyingly long footer}

Just by parsing the number of bulletpoints, and the length of each bullet point (and the first word) I would get a surprisingly good grasp on how things were going, and what was hard/complicated (longer bullet point -> more complex), and very easy to read about exactly I wanted to know.

This is how the so called sutras are written: a page of text gets compressed into one short sentence, so you have to stop after each of them and spend an hour unpacking its meaning, but the entire book is often just 200 sentences.

.zip format in history.

Sutras like Panini's Grammer written 5th century BC are great product of formal knowledge from India. However, writing emails is not the same thing:

- Sutras solved the problem to retaining large chunks of knowledge.

- Sutras had formal rules to keep the meaning unambiguous.

- Sutras were written by the sages who had lot of knowledge and control over language.

-Sutras were written in Sanskrit. A highly malleable (due to SANDHI and SAMAS) yet a precise language.

- English is not the right language to write tersely.

I'm the same and agree with you, but at the same time I've encountered issues with staff who disagree and who feel slighted or depersonalized by less "fluffy" prose. I think you really have to know your audience and Postel's Law applies to an extent.

You have to adapt while writing: If I want a "more important" sentence to be read, I put on it's own paragraph.

This short email style is done to encourage conversations with users to find better market fit.

With colleagues I guess a style that helps to cut down on questions like the one you described is inferior.

I like the idea of this product but

(1) sometimes a BIT of a personal touch goes a long way

(2) even in the examples you give it misses potentially important context. Eg In the board message example, it removes a reminder of the topics. How could you know those weren't necessary? Why is the number of members what it included?

Just doesn't seem ready for prime-time for this use.

My preference would be having it suggest shorter versions of each paragraph inline or something.

For anyone curious, the summary of this comment was:

Just doesn't seem ready for prime-time for this use. I like the idea of this product but sometimes a BIT of a personal touch goes a long way. My preference would be having it suggest shorter versions of each paragraph inline or something. Eg In the board message example, it removes a reminder of the topics. How could you know those weren't necessary?

That's just wrong. The “Eg” is associated with the wrong context, etc..

Additionally, it removed the "helpful resources" from the onboarding email. Though those links in particular weren't helpful, what if it had been actual onboarding links one might send in an onboarding email?

Agreed, some people really don't like terse emails.

I actually got myself into trouble a couple of years back because of this. I've been working with Norwegians almost my whole working life, and am used to being terse, direct and honest. Aside from that, I'm an extrovert, and don't talk unless I have something to say.

Anyway, then I got placed as tech lead for a UK project, working remotely. Within a week, the PM had made a complaint to my manager that I was being glib and not listening to other points of view.

I had no clue where any of this had come from, and sat down with the PM to try to understand - if I'd given the wrong impression, I wanted to fix it. Anyway, we looked back through piles of emails, and in every single case it was a misunderstanding of my real intentions that was directly linked to terseness.

From then on I've tried to gauge my audience better - always using salutations, using longer sentences to say the same thing, trying to be softer etc. I think by and large this has been successful, although I am finding recently my emails tend to be too long...

Why do you have to change and not them? Especially if you sat down with them and discussed it?

Because unfortunately the world doesn't revolve around me :)

It wasn't only the PM, the AI lead apparently felt the same, although I think the AI lead was the one that kicked up a fuss in the first place.

At first I was really pissed off - I didn't think I'd done anything wrong, and was completely baffled about the complaint. I went over my emails myself, and thought about calls we'd had, and I just didn't get it. TBH, I still don't understand how it could have been misconstrued, but I have to accept that it was.

Anyway, I guess the point is that while I'm terse, open and honest, not everyone else is. Consider your recipients before sending that mail - who are they, where are they from, what approach are they likely to be receptive to.

>"Anyway, I guess ... to be receptive to."

In other words, developing your communication skills can have good impact.

How would you decide who would have to "change"?

Presumably, both parties made an effort to better understand each other. There's no evidence to doubt that the PM was more forgiving/understanding of terse emails going forwards.

Plus, OP was new to the UK team. It seems natural that the onus would be more on him to adapt than trying to play with the culture of the existing team.

Am I misreading your comment or something? It comes across as very...standoffish...to me as it seems like the way this was resolved is the most sane, rational, mutually-respectuful course one could hope for.

If the user met with PM and explained the misunderstanding, why do they still need to modify their writing? Is the PM still misunderstanding? I didn't see any evidence that the PM changed here, that's all.

Typically the person receiving the message is better fit to accommodate others (see the HN guidelines about taking the most charitable interpretation of a post). Otherwise you're expecting everyone else to adjust themselves when talking to you.

Have you tested this in marketing emails? that seems to be the profitable marketplace. Even if you can prove 5% increase in roi that's worth a lot to a lot of people!

There are some interesting studies showing short emails perform better but we haven't found that to be true it seems the message is the biggest factor. Some way to improve the message is even more valuable but also likely a much harder ML task

The original inspiration and example featured on the website was from a blog post on cold "welcome" email copy: https://www.gkogan.co/blog/increase-reply-rates/.

I'd love to test it on more marketing emails, or content marketing -- these seem to be the two applications with the most direct ROI.

I think there is still a way to go here.

For example, the "really" in "really appreciate" isn't useful.

Also the last sentence in the "summarized" version reads: "If for any reason you haven't let, let me know..." I would also that that the "for any reason" is not useful, it is again implied.

You don't need "by now", it is implied.

Yep. It's missed some low-hanging fruit despite being very aggressive at removing other content

It's also deleted the important part of a sentence actually requesting feedback whilst retaining the largely content-free 'We're working hard to make life easier for developers and site owners' clause.

And if it was a human editing, I'd expect them to concatenate 'Thanks for joining' and 'I really appreciate you signing up' into 'Thanks for signing up' long before they started deleting links...

I think there's massive potential in the idea, but I'd hope at least the demo video could show something which is more tersely written and easier to read, rather than something which is merely shorter and less useful. That said, I might install it anyway since I can always reject edits I don't like.

Why does it seem common for people to knock modifiers for degree/intensity with words.

The statements "That's nice." and "That's really, truly nice." give off very different messages.

One of them I'm not even sure if you're being short or dismissive/not interested, and I have to do mental gymnastics to figure out what the underlying tone is.

Just one or two words makes a massive (see what I did there) difference in tone + perception.

This is half the reason why it's hard working remotely. I put so much effort into being overly polite/nice with tone because you lose out on all other social cues and it's easy to mistake an ambiguous message.

I'm reminded of the recent post on HN showing that long, personalized emails reduced email conversions, and that short, pinpointed emails increased conversion. Long emails give the impression that one is either a program or desperate, and no one takes the time to read the entirety of every email they receive.

Author here.

If you're talking about https://www.gkogan.co/blog/increase-reply-rates/, this was a primary inspiration for this project -- I even stole the name ;)

I am, here's the recent discussion: https://news.ycombinator.com/item?id=23939462

What about Read Less?

Would be cool to use this tool to read less of incoming email and then dig deeper if I think it's worth it.

I love how you flipped that so quickly, and it immediately sounds more useful to me!

When I was researching this product area, I found a bunch of similar tools that summarized on the inbound side. I thought it would be more interesting to experiment on the outbound side as the main initial focus, which I haven't seen.

I do have a generic popup that you can to summarize any passage, so you could copy-paste some long passage (email, article) to piggy back on the functionality. But it isn't tightly integrated into the Gmail UI as of now.

The demo gif is too long. Gave me the feeling that this product would force me to waste a lot of time waiting for it to load

Based on some digging around, it looks like the app is made with Facebook AI's BART model [1]. The only summarization implementation I spotted was made by huggingface (of course). [2]

[1] https://github.com/pytorch/fairseq/tree/master/examples/bart

[2] https://huggingface.co/facebook/bart-large-cnn

Yes, you are correct -- I linked to this research on the frontpage as a citation.

I see the biggest opportunity here in summarizing books and blog posts. At least for me.

Too often, after reading a blog post, I feel like it was a waste of time. I'd prefer to read a super dense summary first. Then optionally select longer variants before reading the full thing.

Maybe There could be a compress function [1, 2, 4, 8, 16, 32, 64, ...]. Choose the compression level before reading.

Especially with books. I'm not sure how well this could work, but I'd prefer to read 100 books that way first, before deciding on the few I actually want to read.

I'd totally pay for this.

The issue with efficient summarization is the necessity of context. For a message to be effectively summarized you require a bit of "theory of mind"[0], you need to have at least a decent idea of what the receiver already knows. This is something that, especially when done for a single message at a time (with no other info), likely doesn't have a global solution.

Edit: I should be clearer, I believe that the solution shown here does not have this capability, and the effort of using ML to do message summarization is flawed in the general case.

[0] https://en.wikipedia.org/wiki/Theory_of_mind

Your edit and summarization is quite clear. Thank you for that.

I'm excited to see where this goes -- both this product and the general idea of using ML to reduce noise in interpersonal communications.

A human can read the shorter summary and add personal touches as they wish, but the first pass of what the model considers to be extraneous is interesting by itself.

I just put in a 300 word personal email and it spit out a 60 word summary that actually did extract the one concrete piece of news I was sharing, and just cut everything else. That's not precisely what I wanted in this case, but it wasn't a bad guess at all.

Side note: just finished a rewatch of Silicon Valley, I like that the sample text is from someone at Hooli, I wonder if it will change my .... to ... :)

So long as it's not doing that while operating on ciphertext.

Neat idea, I could definitely be more terse in my comms.

As an FYI, some bad English on the Hooli Onboarding Email example:

    If for any reason you haven't let, let me know and I'll make sure you get them.

Good looking out! Will fix that.

Sounds like a simple typo, s/let/yet/

Short emails tend to be either notifications, highly transactional (web based platforms are better for these as they can do tracking and trigger other outside actions), vague, or assume outside context.

These are the same problems as always-on corporate instant messaging platforms, just slightly improved by the cross-organizational functionality and store-and-forward.

IMHO organizations should strive to make email less like instant messaging and more like letters. Fully formed thoughts distributed less frequently and with more purpose.

In short, tax hassling others.

Interesting take because there is a school of thought that comes to the opposite conclusion: emails should be more like text messages: http://five.sentenc.es/.

I think the most interesting part of this problem is the ambiguity in desired result. Emails are inherently unstructured text data, which results in conflict about how people want them to look like.

Seriously. I already write short mails and put bullet points which makeout much smaller. This year one of things I want to do is write big emails like paragraphs atleast two.

I've never written an email that was too long and then later realized it should have been way more concise. I generally know my intentions prior to writing the email, so if it needs to be short, I'll write it short. The reverse situation probably happens more often though (i.e I write a short email, realize it doesn't make sense due to lack of details, then I elaborate).

Interesting. I have a different experience than you -- I often find myself writing an email that's way too long and trimming it down later.

For the design of this tool, it was important for me to "not get in the way" if you're already writing a short email. That's why I designed the widget so that it will only appear in your gmail window if you go over 100 words. It can serve solely as a "reminder" for the rare cases you might go over... almost like a training to help you form short email habits over time.

> I've never written an email that was too long and then later realized it should have been way more concise.

I kind of think we'd have to poll the recipients, not the author, for that sort of analysis.

I find myself trimming down lengthy emails frequently. Not because the information is excessive or irrelevant, but because I know most people in most contexts tend to not read or engage with lengthy emails. Myself included.

I intentionally write short, concise emails.

It has actually backfired on me in the past: because I didn't define and explain the entire space of the matter at hand, I was overcome by other arguments or ideas that I'd already considered and dismissed. That led to followup discussions that didn't need to occur.

It seems that context matters, sometimes long-form is better.

I tend to write longer e-mails - but I agree with you, long-form and short-form e-mails are both optimal for different situations.

The problem seems to be that it's a judgement call - and I'm not exactly sure of all factors/variables that are involved in that judgement call. The subject at hand, the audience, the depth of the e-mail thread to which you're replying, time-sensitivity of the subject, etc. - they all play a role. But my default position tends to be that I'll save time explaining in future e-mails by providing all relevant details at hand now.

Super cool. Would add this as a company gmail feature.

However, I also write emails to reduce time spent in conversation, so while I have learned to be more terse, when I'm not, it's to provide references so that I can shorten or preclude a conversation.

An ML email shortener is a fantastic idea for business comms, with the caveat that short email culture can also reward vagueness.

Shortness shouldn't be the goal. Clarity should be the goal. The problem is that clarity is a much harder thing to test for.

I would pay money for a tool that checks my technical writing for reading level and warns me when passages go above Nth grade. Most people in the US graduate high school with an 8th grade reading level, and some people may not be able to read above a 5th grade level. That means tiny words and short sentences are king.

There are also metrics for clarity and simplicity that proxy on number of syllables. Those are really useful and don't require a complex ML model.

https://readable.com/ is an interesting tool I came across in my research you might be interested in. It relies on the Flesch–Kincaid readability tests [1] and has found sustained usage.

Actually, now that I think about it... this might be a good source for bootstrapping further training data.

[1] https://en.wikipedia.org/wiki/Flesch%E2%80%93Kincaid_readabi...

Coincidentally I've been feeling like my emails are becoming way too verbose recently. I'm not sure what it is, but I'm just taking longer to reply in a way that conveys my message properly, or I'm finding myself editing a draft to death before I ever send it. I used to be able to just reply and move it to the next bucket.

Can I use this in conversations with my wife? So much talking. ;)

Joking aside - this is a great tool in corporate settings. In technology, one works with two types: Very verbose (tend to be introvert or detail oriented) or Too abbreviated. I cannot stand reading detailed accounts - and this could add to many professionals comprehending more.

If this could convert passive voice to active voice, or underline it as an error, this could be a great technical writing tool for inexperienced writers. Good technical writing is hard, a tool similar to this could be useful.

I wonder how this would perform compared to using GPT-3

I just got access to GPT3! Have only played around with it for a few hours so far, but haven't gotten it to reliably summarize.

Nice job doing the last mile and making it a usable service.

Just guessing: SMMRY or Quillbot with a little enhancement to preserve the salutation?

Nope, I've developed and hosted my own summarization API here. It does heavily rely on open-source machine learning libraries and research / datasets released by Facebook AI [1]

[1] https://ai.facebook.com/research/publications/bart-denoising...

Reply on news.ycombinator.com