I Received an AI Email

> Could an AI agent craft compelling emails that would capture people's attention and drive engagement, all while maintaining a level of personalization that feels human? I decided to find out.

> The real hurdle was ensuring the emails seemed genuinely personalized and not spammy. I knew that if recipients detected even a whiff of a generic, mass-produced message, they'd tune out immediately.

> Incredibly, not a single recipient seemed to detect that the emails were AI-generated.

https://www.wisp.blog/blog/how-i-use-ai-agents-to-send-1000-...

The technical part surprised me: they string together multiple LLMs which do all the work. It's a shame the author's passions are directed towards AI slop-email spam, all for capturing attention and driving engagement.

How much of our societal progress and collective thought and innovation has gone to capturing attention and driving up engagement, I wonder.

As for the humans, we went fishing instead.

The author sounds unfamiliar with this brand of marketing email, so I can see why it would come off disquieting to find it’s all AI - but it’s equally annoying from a human.

At least with AI sending this crap nobody can use these emails to justify their sales bonus.

If the dead internet theory isn't already true, it is going to be soon.

Such "personalized" cold outreach is seen as the next holy grail by marketers and will be a common sight on LinkedIn, Twitter, Email etc, soon.

You could refine this in further iterations by also adding examples based on previous correct/incorrect interest predictions, thereby effectively reducing the amount of spam / making cold outreach suck less.

There are different ways to use AI to achieve the same goals, some more responsible than others.

I've actually made an internal company April fools website. Too bad I've never kept a copy but here goes.

It's called Proxy Ai. It reads your emails so you don't have to. It reads every posts on social media so you don't FOMO. It communicates with those chatty colleagues so you don't have to. Proxy Ai... So you don't have to.

"That actually sounds like a pretty good product. Does it send you a summary of the conversations, emails and social media posts?"

"No"

It said:

Hi -

Just a note to say I'm a big fan of your writing. I always learn something and love your voice, which is hilarious and singular.

Write a book!

Best,

{Name}

{Link to sender's startup}

{Link to sender's substack}

New to writing online, it made me feel really good that someone enjoyed what I wrote and took the time to write and say so.

After reading this piece, though, I went back and read it again, and I just don't know. It's not quite GPT's usual voice, but it is strangely non-specific.

The startup is an AI startup, the person's Substack is full of generative AI illustrations, and they do seem like an AI fan, but reading their posts, they also seem like someone who's genuinely interested in preventing a dystopia.

I suppose receiving encouraging emails from strangers is just another situation that'll have us looking over our shoulders now, on guard, trying to walk the line between naivety and paranoia.

If I ever receive spam addressed to foobar.com@mydomain.com that is unrelated to your service I know you leaked or abused my data. Result: you will get a DSGVO complaint and I filter all emails addressed to this address from my inbox.

The good thing about using a catchall email address is that I don't have to create a mailbox for each service/purpose, I can just make email addresses up as I go. All you need for that is your own domain and a mailserver that aupports it.

I never fell for a spam mail so far (i.e. not once clicked a link like OP did), but I fully expect this will change soon. Tough times for people that commonly expect mail from random strangers.

Hadn't touched marketing for ~5 years, as I said I know the org well so I thought it will take me about a month to get the next 6 months of marketing built and automated. How wrong was I. 7 days later, the full marketing org is running, at a decent scale, on autopilot, for a year, and I don't know if/when I'd need to hire someone into marketing.

Marketing has not fundamentally changed, but it's changed such that one individual could fully operate the fundamentals. Personally I love it, I'm sure others are going nuts.

> Hey Raymond,

Thank you so much for your kind words about my post on revamping my homelab! It’s always a pleasure to hear from someone who appreciates the journey of continuous improvement. Your message truly brightened my day.

Indeed, using Deno Fresh for my blog has been an exciting adventure. The process of managing updates and deployments, while sometimes challenging, has been incredibly rewarding. It’s like tending to a garden, where each update is a new seed planted, and every deployment is a blossom of progress. The satisfaction of seeing everything come together is unparalleled.

Your introduction of Wisp has certainly piqued my interest. A CMS that simplifies content management sounds like a dream come true, especially for someone like me who is always looking for ways to streamline processes and enhance efficiency. The name “Wisp” itself evokes a sense of lightness and ease, which is exactly what one hopes for in a content management system.

I would love to learn more about Wisp and how it could potentially fit into my workflow. The idea of having a tool that can make content management more intuitive and less time-consuming is very appealing. Could you share more details about its features and how it stands out from other CMS options? I’m particularly interested in how it handles updates and deployments, as these are crucial aspects for me.

Thank you again for reaching out and for thinking of me. I’m looking forward to hearing more about Wisp and exploring the possibilities it offers. Let’s continue this conversation and see where it leads!

Best regards, Tim

No one believes the CEO has taken the time to email you with onboarding instructions immediately after signing up anymore. But outreach tactics like this are still quite manipulative.

This person wants me to buy their product, and before they can get a word out about it they’re already lying to me - about the origin, the intent, the faux thoughtfulness.

I want nothing to do with shameless dishonesty. This isn’t the way to sell your product.

Wisp, if you’re reading this, I now have a permanent negative image of your brand.

Enormous amounts of email will be generated but no one will ever see it.

---

Hey Travis,

Checked out the Next.js Notion Starter Kit. Amazing project!

Noticed you might be juggling multiple tools to manage content. Ever thought about a headless CMS that can streamline this?

Wisp might be a handy solution. Let me know what you think!

Cheers, Raymond

>> Have you ever received an email that felt so personalized, so tailored to your interests and experiences, that you couldn't help but be intrigued? What if I told you that email wasn't crafted by a human, but by an artificial intelligence (AI) agent?

> I don't really have words for this, but I dislike this.

What a classy understatement. I find the strategy employed by Wisp predictable and infuriating. Like insects or other near-automata, humanity is racing to the bottom with "Generative AI". And I use "AI" in the loosest possible sense here, because once you pull back the curtain, current tech is actually only a slightly better Markov chain.

After using chatgpt regularly, it's responses to anything but the most trivial, clueless questions are riddled with errors and "hallucinations". I often don't bother anymore, because it's easier to go to the original source: stackoverflow, reddit, and community forums. Gag. It does still make a good shrink / Eliza replacement.

AI spam emails will definitely happen on a large scale in the future, but on the optimistic side, we'll have AI assistants to read every email. Personalization won't really matter; it only matters if the marketed service is genuinely useful to me. In the future, if I have a need, my AI will find the best solution by considering many options. Previously, this was time-consuming, but now AI ensures we find the most suitable solution.

No fancy marketing will be needed because AI will filter most of them. In the end, marketers will find that the most efficient way to market is to honestly list out your service/product specs, as AI will compare them. On the other hand, for things I'm not sure I need, AI will help judge if they are indeed useful to me, regardless of how fancy the email/call is. If they are, it will facilitate cooperation; if not, it will skip them.

Therefore, marketing may still have a role: to help you discover things you aren't fully aware you need, and AI will help you decide if you really need them.

>"I knew that if recipients detected even a whiff of a generic, mass-produced message, they'd tune out immediately."

Then don't brag about it on your blog! Sheez.

(Ok, so technically he's not bragging about it on his blog, because it's probably just an LLM bragging about it on his blog for him, but that's the point!)

> Does this mean that I should private my GitHub-mirror to my personal blog, because this can become a common thing?

Abusing public information on GitHub has become more common. The other day, I received some cryptocurrency spam ads from GitHub. It turns out to be a bot injecting ads as issues on other people's repos and randomly @ing accounts. It deleted such issues immediately, so the net effect is that I get an unfilterable spam email.

Did he use an LLM to write the blog post too?

> It felt like a family fridge decorated with printed stock art of children’s drawings.

Yep. "Generative AI" is like an infinite clip-art gallery that can be searched with very specific queries.

The coin has two sides: in some situations it devalues human effort - as in writing (long/detailed) documents in formal language is now attainable by everyone. In situations where sincerity and originality matters, human effort has now increased in value.

Watch out recruiters, AI can do better than you! Not like I will like these unsolicited outreaches more, the exact number is zero, how many times I found these useful or relevant before when biorobots wrote and sent and administrated it in half or just few minutes, and I do not look forward having these now on mass scale when hundreds of AI could write thousands, flooding my email account, making it absolutely unusable.

Just as most of us ignore calls from unknown numbers, we may also default to ignoring emails from unknown senders in the future. This could lead to a reluctance to send emails, as they might be perceived as "unknown" to the recipient.

Whether they are AI or not, I have no idea, but sometimes, and recently in emails, I purposely make a typo or grammar mistake to add some "human" touch to it, knowing that an AI will always type a perfect one.

How it was written is not relevant. Off to the trash it goes.

> At the same time, we need to establish guidelines around transparency and consent for AI-driven communications at scale. Deception through omission is still deception – people should be aware when they're interacting with an AI agent versus a human.

This is clearly pissing in the pool. I've gotten so much value from people who have made their emails public with a 'if you're curious or learning feel free to email me' (e.g. patio11) and I've long had the invitation in my HN profile too.

Nasty for people to abuse this to extract value for the few weeks/months it takes people to realise what's happening and make themselves harder to contact.

This reminds me of AI-generated fake security vulnerability reports about curl: https://news.ycombinator.com/item?id=38845878

They reached out to me, asking whether my company would be interested in Something Somethingification. I decided that since I don't even understand the term, I'm not the right person, and decided to ignore it.

Then they followed up. Meh.

Then they followed up again, and I thought "okay, a little reward for perseverance", and replied something along the lines of (I don't work there anymore, no access to the original):

"Hey, thank you for reaching out.

Unfortunately, since I don't even know what Something Somethingification is, I am not the right person to talk to. So I'll kindly pass and consider this email human-generated spam. Thanks!"

A response came. Within a minute, barely seconds after "undo send" disappeared.

"Who would be the best person to reach out to, then?

By the way, this is a GPT assisted conversation, so it's a computer generated spam."

WHAAAAT. This really got me. Remember, it was 2021.

"Okay", I replied, "Now you got my interest!

How many such conversations are you able to have at the same time?"

It replied, within a minute. It contained a quite from Arthur C. Clarke that "every technology advanced enough is indistinguishable from magic" and his picture. And an answer: "Actually, sourcing contacts is the bottleneck, so we have only a few of these each day. Anyway, do you happen to know who we could reach out to instead?".

I was amazed, I decided I'll reward this with what they want.

I replied how impressive it is again, as the whole conversation made sense, and it gave them a contact to a director that could be the right person. They won this one.

We need to update our spam filtering techniques, fast. Somehow. But how?

They will all lose money, time and more with the coming wave of spam and fraud.

"Hey, love your work. random flattery What do you think about mine?"

I've received a few messages like that before LLMs were around, just an annoying self-marketing technique.

Something like a marriage of a digital signature with a captcha: the message has a digital signature of the sender that can be verified with their public key, but it is somehow verifiable that the particular signature provider only does the signature if a human being completes the (difficult AI-proof) captcha.

Something like this approach can at least mitigate the mass AI email problem, although the one-off AI emails are unlikely to be slowed by this approach.

It's possible to use a noreply.github.com linked to your username for making commits. And you can to change the authorship of past commits in your own repos with write access.

I try to avoid give my email in a public and processable format whenever possible.

The only problem is that they referenced a role at a company I'm no longer at. The, presumably AI, author crafted the email in reference to my former role at a different startup.

After seeing this thread, I decided to follow up on my AI suspicions. Nothing conclusive, but that person is currently touting that they've sold their "course" to "1000+ founders."

No thanks.

Both are unsolicited emails, i.e. spam.

I feel confident that Gmail’s spam filter will be able to handle this quite well.

I’m betting that the introduction of LLMs will not change the fundamentals of spam-fighting.

https://paulgraham.com/spam.html

https://docs.github.com/en/account-and-profile/setting-up-an...

I'm not after shallow interactions today and I would use it (much like a dumb spam filter) to judge a new sender's respect for my time expecting them to have stated their business with total upfront clarity, not mystery.

Everyone's spam filter is tuned differently from others', so spammers had a hard time beating this with automated messages. About the best they could do was adding random keywords in hopes of triggering someone's positive "not spam" trigger.

Now spammers gain personalisation at scale, so this advantage is at risk.

And also from the About page on the linked website

Even now, we're starting to have a sense for which images and text were AI generated. And they'll evolve to get around the antibodies. And we'll build new ones.

https://github.com/skorokithakis/spamgpt

It was a bit of fun, until I realized that most of the replied from the spammers were AI as well. We were just automatically spamming each other while OpenAI made money.

I stopped using it then.

Although he got more click-throughs to the top of his funnel, none of them are going to pass through to a conversion because once you reach his site, you realize that he's deceived you.

That he doesn't even realize this is concerning...

The general public doesn’t want or need it. They want to work less and get paid more.

Maybe in future I will have my ”AI secretary” to answer those and have a discussion with the ”AI sales assistant”.

I talked to many people, and all have developed immunity against the cold outreach.

You definitely should mark this email as spam so this cannot become a common thing.

People sending AI crap to others should have their email accounts banned.

One issue I see is that it’s much harder to employ an LLM defensively (for filtering) than offensively.

Welp.

Specifically smart filter to remove SPAM in a smarter way.

Most people get a lot of spam from sales agents, SEO services, start-up accelerator, etc...

With GabrielAI you can say stuff like:

"If the email is from a SEO agency or it is trying to sell me SEO service"

Then move it to SPAM.

Similarly for all other type of spam or emails.

You can also move stuff to different labels in Gmail to organise your inbox.

Spam is spam?

The "upside" is that nature eventually takes care of things when they go out of equilibrium, so there might be a forest fire on the horizon to restore it. In the case of AI spam, it might cause people to automatically filter their incoming mail from any content that even implicitly tries to sell something, or even any email arriving from an address that is not on their whitelist. This might eventually cause people to need to actually physically meet (gasp!) in order to add each other to their whitelist.

Edit: "Unnecessary" might be my judgement, instead of "acceptable."

> There's also the question of ethical considerations around using AI for mass personalized outreach. While my experiment yielded positive results, with recipients appreciating the personalized touch, there's a potential slippery slope.

Unbelievable... I'm not a philosopher, but in my understanding, being ethical doesn't mean walking the line just fine so as people don't call you out on your bullshit.

The ethics of an action is of consideration both BEFORE and after executing it, and on the merit of the action itself!

Cold spamming is illegal where I'm at, probably Europe as a whole?

...

2024: AI impersonating Bill Gates sends you SPAM

It’s sad that going forward I probably won’t be able to tell genuine interest from this kind of fake bullshit.

This will make it worse.

Solutions? At least some could involve key exchange. How about a bounty of some sort on spammers?

... shall we tell him?

Dug?

Guys, it’s a tool like any other.