Show HN: GPT-3 powered service thats helps you send more humane emails | Hacker News Reader

Show HN: GPT-3 powered service thats helps you send more humane emails

remarkai.in

35 pointsylankgz35 comments

You are joking right?

You know what makes a more humane email? A person sitting down and writing that email like a human being.

This, of course, isn't just an issue with tools but also marketers who popularized this nonsense. Brian Dean, Neil Patel, just to name a few.

These emails _stink_ and if you are comfortable with leaving a bad impression on people then by all means go ahead and write this gibberish.

And yes, I'm self-aware of how critical I am being but unfortunately people have become numb to emotion and will rather send 100 automated emails for a chance of 1 conversion, than a handful of personally written ones for a chance of having them all lead to something.

On top of that, you're wasting people's time.

These emails that have the buzzword of "stumbled", "your audience", "loved your content BUT can you do THIS for me? - get instantly sent to Trash.

How low do you have to stoop to be comfortable with sending an automated email that says, "I am impressed by what you created.".

Are you, impressed?

This response would have benefited from a more humane wording. I wish the tone on HN would be less hostile. Just shows that some help wording things can be a useful tool.

The effort reeks of automated hypocrisy.

I am aware that hypocrisy is tolerated in vast areas, but be reminded that it is a mortal offence in others.

A "kind" word has the opposite value when said unfelt.

People are not great at detecting hypocrisy from people they don't already know intimately. The general idea of automating pleasant sentiment in personal async communication should work (not this implementation, which seems more fit for mass mailing) for a bit. We will be quick to adjust to recognize it as automated and see roughness or even borderline rudeness as a positive signal of personal attention.

> People are not great at detecting hypocrisy from people they don't already know intimately.

I immediately suspect hypocrisy when someone I don't know sends me a super-friendly, over-familiar, and flattering email asking for something.

I mean none of that. Just nice polite and positive-sounding messages. For me that simply describes most of communication except with close friends/family (there were a couple of colleagues who would swear and be negative or frank to the point of rudeness but it's generally rare).

I'm sure in most cases people do not genuinely feel that way towards me and are faking niceness (being hypocritical) but rejecting all of it would mean rejecting a lot of actually nice people, so with people I don't know well I assume good faith and respond or not based on the substance of what they write. This is what I mean by "bad at detecting hypocrisy".

Nobody is rejecting niceness.

We are rejecting falsity - it is lying. We are warning: don't do it. Act properly. Properly is "respectfully". With proportionate measures. Lying is plain disrespectful, and more.

Being nice and warm to people is lying and falsity unless you feel that way towards them genuinely. If you don't think so then I guess you may be among those employing such hypocrisy without a second thought. In routine communication I probably won't detect your hypocrisy because again people are bad at this. I will assume you are genuine and reply to you. An idea like proposed, if properly implemented, may help other people who are less adjusted and perhaps incapable of employing such hypocrisy.

I am suspecting that you are holding assumptions which are not the case. So, let us be clear:

-- yes, you are supposed to be «nice and warm» only in case this is genuine. If it is not, just behave appropriately. "Respectful" does not imply «nice and warm». Also: since «nice and warm» does not imply "respectful", and on the contrary it may suggest the opposite if patently constructed, I would be very careful about any temptation to think that some coating "will do";

-- when suggesting the idea of people «employing... hypocrisy», it is very advisable to avoid the term 'you';

-- teaching people hypocrisy, idea which may appear from your text, would be criminal. If you had to orient them, it'd better be towards the substance.

And I recommend to not use 'we' when speaking for yourself.

Regarding your nice -> respectful update, just replace 'nice' with 'respectful' so the proposed idea reads 'correct emails to be respectful if they're not', done.

'We' is perfectly valid if not using an individual matter but a sharable position: "I went there", but "We believe". (Because it is not just this one who happens to be writing.)

(By the way, I could add for clarity, there is an extreme difference from the considerations that are apparent in the case of that 'we' and the suggestion "Who has never set fire to a public monument? I did" (Graham Chapman) that was implied in that 'you'. It is really not the same, because avoiding calling people "arsonists" is what I recommended.)

The idea to have engines that notify of disrespectful qualities in text is interesting - of course it should be clear that it is a completely different matter from what was discussed before -, but I suspect that the engine should be quite intelligent to discriminate what is respectful and what is not. It is similar to the former context: a veneer of kindness is cheap and possibly inadvisable; a respectful stance is not superficial hence not cheap - it has harder requirements for feasibility.

Real example: https://twitter.com/DannyRichman/status/1598254671591723008

> Because it is not just this one who happens to be writing.

I am talking to you and you are representing only yourself, not a collective.

> some help wording

And which wording would you have suggested, to improve the literal post from Skilled, while retaining exactly the meaning Skilled intended?

And systems similar to the one submitted would do _that_?

Why? They're using big banks of computers to try and emotionally manipulate people at scale, do you really think that being nicey-nice to them is going to bring them to some sort of moral epiphany? I tried it for years, they just classify that as sucker behavior and laugh it off.

Doesn't seem to have done Salesforce any harm.

But such tools may help people overcome or reduce tension from cultural differences. As a Scandinavian, I prefer to be polite but very straight to the point in my emails. When communicating with some foreigners, their style makes me feel cold or impolite and I have to use a lot of time and effort to carefully word my emails more "humane". It feels awkward, unnecessary, and tiring, but I really don't want to offend anyone. Getting some help to master that style would be very helpful

I think you're right to be critical in this case, and I'm not defending the service. But as far as whether you can use GPT to write a more humane email, you definitely can. You just need a well-constructed prompt.

At it's core it's just a text prediction machine. If you prime it with language such as, "the recipient of the following email found it to be thoughtful and enjoyed reading it. The recipient was amenable to what the email said, and had a good opinion of the email's author afterwards", it will try to predict a message that would most likely follow that kind of text. I have found that it's usually better than myself at accomplishing that.

Isn’t big part of what makes something humane that it comes from a human being, who means what they’re saying?

There’s something uncomfortable about trying to simulate ‘the human touch’ to improve conversion rates. When the mask slips, and people discover your emails for what they are, I’d say it’s more alienating than any traditional mass email

Mass marketing emails are one thing. On the other hand if you believe that you sometimes rub people the wrong way despite having the best intentions, then you could be using it as a genuine attempt to communicate empathetically with the other person.

Not everyone is an equally effective communicator. Plenty of people struggle to strike the right tone in written communication, others aren't even aware they're missing the mark.

A tool like this can help people ensure that the message they're sending does in fact match their intent.

Gross. This is the digital equivalent of those awful junk emails with fake handwriting printed on the envelope/letter. I automatically reject anything that comes at me this way since the opening move in the communication was a lie by the sender trying to create a false sense of personal communication. Detecting AI in text email is a lot more difficult but the discovery that a firm was using this approach for its marketing would be an instant blacklisting from me.

I know your intention is to help others grow their business and you've doubtless invested a lot of work in this. But since the ultimate purpose of your tool is to automate the business of persuading people to spend money, please stay the fuck out of my inbox.

Before looking at the post, I thought it might be useful in this age of remote work where tone (that rarely exists but is imagined/inferred) is more important than ever. Disappointed to see:

> The best email templates suggested by AI to help you ace your email marketing and increase conversions.

So it's for 'you' meaning companies (not people) to seem more human.

Another reminder that sending mass email should not be nearly free in our shared internet infrastructure.

These examples look pretty close to the form letters I get from recruiters on LinkedIn. Because let's be honest, it's not possible to send a personalized cold sales email because you don't know the person. Trying to be overly familiar in an initial email is a turnoff to me in any context.

I was assuming this would be about filling in the blanks on emails you share with family to keep in touch or something, but that's a harder problem since then the AI needs to actually know things about the relationship. Still, certainly seems like an approachable problem by tweaking some novel-writing AI to consume your personal emails and then use an interface like NovelAI or something to co-write an email with the AI (or KoboldAI to keep things local).

> filling in the blanks

I., should not we solve real problems? Did we run out of real problems last night as I was sleeping (thank you for the good news!)?

"Real problems": when I compose, e.g. these posts, I verify not only the thought structure but also the terminology, and often check Etymonline when I am not fully competent on the original intended meaning of a specific term. Such operation, legitimate, could have ways to be facilitated through automation. I do not feel a need for it, but it is an example of a possible area of tools for an improvement of processes.

Automated manipulation - there exist facial expressions appropriate to approaching that.

(And look, you would not want to receive false tones from a firm: communications with close ones has orders of magnitude higher requirement for genuinity.)

Edit: and already if you notice the silent turd hurling that I see hit this post in these minutes as I am writing, you should get the idea that there do exist real problems. To ignore that would be decadence. We should remain focused - while not denying that play has a role in development; play remains functional to the development.

Taking a cue from the provided examples, the intended usage of this is for sending prettier spam messages. So basically this looks like trying to put lipstick on pigs. I'm not sure about others, but i would still NOT be impressed or appreciated if i received more "humane" spam...

And now I’m thinking about how to build a spam filter that runs parts of emails through GPT and then matches the output with the entire email to arrive at a score.

Because I would love a way to automatically filter out generated bs like this. Hell, even have it interact with the sender totally automatically to waste as much of their time as possible!

Not only in Emails but also auto-generated blogs. I really don't want to waste time on some generated content and realizing later it simply doesn't work.

Apart from the whole I don't think that word means what you think it does thing, the issue with this is that the output seems to be equivalent to what you would expect an American marketer to produce and spam out after scraping linkedin. Presumably this is because there are a lot of examples of that style of email present in the training data. Personally if I received any of these I wouldn't even get past the first line before marking it as spam, which probably wouldn't do the reputation of your mail server much good.

It's an interesting, if slightly evil, idea, but I'd want the style of the output to be different or customizable before actually using something like this. In other words it needs a "tone of voice" setting. Write intro to someone in voice of cynical Brit / enthusiastic Canadian / morose Finn etc.

A no BS german tone of voice would go a long way

Let our AI help you communicate like a human being! It's a funny state of affairs when AI is touted as the solution to inhumane communications. But I suppose marketers need all the help they can get to present as human.

Although, I'd suggest that the truly humane action is to stop sending spam emails in the first place.

I think marketing this as more humane isn't the way to go. That's kind of annoyingly untrue. It's a human consulting a AI.

Plus sometimes the AI says things that are untrue too. Sometimes you don't want to tell the other guy "I love your work" even if it gets attention because they might ask which my favorite was. So I don't think the output is necessarily ready to send out.

What this is really good for is to help create a rough draft. It allows the user to create more wordy, but polite, copy that the user can then edit to their own voice. Or you know something like that.

Sadly, AI-driven SPAM like this has pretty bad side-effects for the entire economy because it erodes trust.

Whenever we have a public job opening, I put an uncommon and unrelated word into the text description and mandate applicants to reference it, like "transparent frog". Any email who doesn't contain the phrase gets deleted immediately. It's my version of a CAPTCHA to get rid of this SPAM.

At the current rate of nlp understanding, I suspect your captcha may need to be updated.

Looks like spam, reads like spam, if you think this is "more humane" then you have never talked to a human.

I'd be dissapointed in my Spam filter if any of these found my eyes.

Reply on news.ycombinator.com