Honest question: Anybody have a use-case for this in mind? I can't remember the last time I read an email for longer than about 5s and didn't fully care about the content. Business emails should already be formatted for efficiency, and I'm going to want to be sure I catch each point of the long-form ones. Maybe vacation updates from family?
Maybe you the user is not the target here. Maybe adding it to yet more things will make many important people at Google even more important and Sundar can babble on about AI first to the shareholders.
It is hard not to assume just this with anything Google does.
Yes, people will send long AI generated emails, then AI will summarize it back.
I can think of one - there's a bunch of newsletters/substacks I subscribe to that occasionally contain stuff I really care about somewhere in them, but are too long and dense to quickly skim them. An AI summary of most important points and topics would help me decide whether to invest time and read it in full, or archive it and forget.
That's all theoretical anyway. I'm not on GMail; I technically have this feature in the e-mail client on my phone, but I never even use it.
EDIT: The article mentions those summaries would apply to email threads too. I think that could be helpful. I've got tons of threads from some mailing lists that tend to grow large (10-20+ messages); catching up with them or revisiting old ones is tedious; an AI TL;DR of such threads actually sounds useful.
Also I'd love for e-mail clients to have a feature reminding normies when they've "forgotten" to address some questions from the sender.
> I can think of one - there's a bunch of newsletters/substacks I subscribe to that occasionally contain stuff I really care about somewhere in them, but are too long and dense to quickly skim them. An AI summary of most important points and topics would help me decide whether to invest time and read it in full, or archive it and forget.
At this point in time though an LLM summary would not be guaranteed to contain all the points or to focus on the stuff you care about, so it wouldn't be reliable anyway.
On the contrary, LLMs have been good enough at summaries for over a year now.
For this task, AI summary doesn't need to be 100% reliable anyway. I'm fine with it missing some points; it's better than me missing all the points. I'm fine with it hallucinating or reporting the opposite of some claim - the very mention of that claim tells me the article talks about it in some way.
Also, it's not like there exists an alternative. You cannot do this task any other way - and in particular, there's no economical way to get humans involved in it.
I don't think people really appreciate that last point. Everyone's quick to complain that LLMs don't give you signed, notarized guarantees of being more thorough and accurate than a team of scholars studying a text for a year; few stop to think that they're already better than average person and few could afford to hire a human to do this job even at worst possible quality.
This question but for every AI feature every company is going for.
I'm not sure at all why, but there seems to be some economic incentive to push for AI features. Is it because it's a buzzword? Is it because selling GPU time gives good margins? Is it because the all knowing efficient market demands it? I don't know, but users don't want it.
It is because if you throw enough things at the wall, some hopefully will stick.
I do see lots of potential AI to improve my workflows, but this ain't it.
I am also very annoyed by google chromes dev tools recent new AI features. Don't need them, Don't want them - but I cannot get really rid of them. They are in the way. And they are in the way, so people use it, so some internal metrics make someone happy. Just not me.
I'm leaving gmail.com and gmail.app because of how hard they push AI with no way for me to opt out. I've been using them for many years.
(I might continue to use my gmail address using POP3 to third-party client software.)
I'm a fairly heavy user of LLM services, but don't see the value of "help me write" or "summarize this" or "happening soon" (Gmail's attempt to insert itself into my relationships with shippers and delivery services).
I’d love to ditch the multiple google mail accounts that I setup for nonprofits, but I worry that google is better at catching spam and fishing than the competitors. I use outlook for another, but MS is insanely unfriendly to newb users and their spam filters don’t work as well. Any suggestions?
I will probably return to Emacs/VM as my client, a very old version (5) of VM specifically. For hackers only.
Using a desktop email client is my preferred way to use email. I can read them even offline.
I'm hard pressed to think of a situation where a desktop app isn't superior to a website. Web apps are a port in a storm; if I can use a desktop app, I always prefer to.
Gmail has no desktop app, though it has mobile apps.
The web or PWA Gmail has more features and settings than any third-party desktop app would offer when using IMAP or POP. Gmail is a custom-designed first-party app. If you read mail from Google through Outlook or Thunderbird, it will have lowest-common-denominator features through the lens of a different vendor's structure.
Now you may consider it a superior advantage if the Gemini AI isn't available, but probably won't be able to create custom filters without the PWA.
I have a rather new experience with customer service: they don't understand my questions, and they don't appear to have read the questions.
My theory is that a lot of companies are starting to use Ai to draft up responses. These Ais don't understand the question, but picks the closest in concept space, and assumes that question.
It is horrible, because I need to all caps and explicitly ask customer service to read my messages, and I need to become incredibly rude to break out of it.
This has been kind of happening in Customer Service for at least a decade now, just not "with AI": You send them an E-mail with a specific and detailed question about your bill, and their inbox triage bot reads it, sees the word "bill" and responds back with a generic "Hi, thank you for contacting us about [BILLING]. Please find information about [YOUR BILL] at this URL: xxxxxxxx"
They're called canned responses. A lot of customer service reps are forced to use canned responses at level 1 no matter what. They can't respond with anything but a pre-made response.
Hope it’s better than Apple Intelligence, which seems to mash all my different emails together in a one sentence summary even from widely different senders.
It's actually wild how terrible Apple Intelligence is. You'd get a better summary by literally copy-pasting the email/text message in any LLaMA model, so I'm not even sure what they're doing to get such awful results.
They just suck at AI. They probably joined the game too late so their in house product is still at the "kinda sucks" phase.
When I tried local llama for this (on a computer), it used tons of battery and memory; I guess it would be hard to use that on mobile device; they probably need to do a lot of work to make it feasible on iOS and locally and it just sucks.
On the positive side, I don't feel bad for having iPhone without Apple Intelligence.
Apple Intelligence had so much promise and so little delivery.
Mine frequently hallucinates (tells me I have a sale for instance on eBay that doesn't exist) or even gives me summaries that are literally the opposite of what's being said (so far innocuous, but I have already imagined how this could have disastrous results in a different context)
Opposite meaning is actually quite common mistake when I tried to summarize e-mails with Llama and ollama. It catches the general themes correctly, just inverts the meanings, making it useless.
I love Apple and almost exclusively purchase their products. That being said, they will never lead in this realm. The worst students from my graduate studies joined Apple and NVIDIA. The top researchers in this field went into Google and OpenAI. Just look at linkedin profiles. Over time, Apple is destined to fall behind, given they do not pay as much as their competitors, and are much less reluctant to invest in the capital required to improve these models (despite larger capital, Apple leadership is surprisingly illiterate. Cook and Ternus are not software guys). Google has TPUs, and I am of the opinion their model will be used in future iphones (Sundar and Sam Altman are other illiterates, but there are very talented people at Google rn and should they have the reigns, I am confident they will dominate this field). Just looking at linkedin, the people that published tiny improvements in models are at Apple, those who truly contributed to these models through publications are at Google.
The recent revelations from The Information and quotes supporting it that came from Gurman last week back up your overall prediction.
Interestingly, Cook is said to be a huge believer in AI, yet he was unable to execute on it, even by making an ideal hire in John Giannandrea. JG toiled at Apple for years, unable to make groundbreaking progress due to Apple's corporate culture. It's an ol' boys club of forceful execs who have worked together for decades, and they all opposed spending what JG wanted.
If JG couldn't make an impact, no one will. What disappoints me the most is that Apple's "Not Invented Here" syndrome is so severe that I suspect they'll never admit full defeat and let users replace the rotten-to-the-core "Siri" assistant with a competent one, no matter how far behind they fall.
Right now it's Google's game to lose. Once Android devices are doing the kinds of things that should be easy for current models to do, real things as useful as a human assistant could do, it'll be the first meaningful challenge to Apple's dominance on the high end of the market. Things like "Answer all my calls while I'm in meetings, tell all sales calls I'm not interested, and for all others, if it isn't an emergency, take a message and tell them I'll call them back, and book 15 minutes on my calendar within the next day to do it."
I agree. I lost hope in Apple's AI "efforts" when I found out that their MLX team (responsible for making a Pytorch/CUDA alternative for Apple Silicon) DOES NOT have access to the source code of ANE (Apple Neural Engine)!! Only the team behind Apple Intelligence uses ANE, not the MLX team.
Talk about fragmentation and lack of trust/hope in your product.
Somehow we ended up doing the opposite thing of compression (and I do not mean decompression): user writes three word email, AI expands it to several paragraphs of useless text, sends it to recipient. There another AI digests all that text to some other three words, loosing original meaning and not only wasting CPU on AI tasks, but wasting network bandwidth by sending a lot of useless data. I hate 21st century.
"for android and ios"
imo the article's rather negative tone is not justified here, when I only have a small screen I think I would welcome an AI summary
in any case I'm willing to see what it looks like before panning it
best case: it is so full of lies no one trusts it to begin with
worst case: it is mostly ok, and only the few emails have opposite summaries. So you are nornally fine, until that "meeting NOT cancelled" email is summarized "meeting is cancelled", and everyone is wondering where you are.
Migrated off of Gmail a few years ago and the gains of not having to worry about Google's direction keeps paying off.
Go with a provider who's actually focused on email, you'll thank yourself.
Personally I truly, deeply hate AI summaries and so I strongly disapprove of Google’s latest moves into this territory. I know it’s not a comment with substance, but I think the more people express their aversion to this trend, the higher the chances of it not becoming the norm.
Yesterday I noticed the "Upgrade", with a persistent (1) badge on it, in the bottom left corner of the web app, trying to get me to buy some plan with AI features, even after I looked at it. I installed Thunderbird immediately. Never touching Google UI ever again.
Keep going. Here's the next step after Thunderbird:
https://stuff.sigvaldason.com/email.html
(Hint: Emacs)
:-)
Thanks for the recommendation, I went ahead and installed Thunderbird too. Tired of this AI shit being shoved everywhere even if it's unwanted and unhelpful. Done using the Gmail app.
Been using this in our Workplace account. Most people don't like it because they are luddites, but I absolutely love the summaries for those multi-week 50 email threads. Also the ability to quickly insert an absurdly overly formal response to a simple request as a joke is amusing.
I do wish it had the ability to respond to emails for me though with a prompt.
Oh please, not everyone who disagrees with you is a luddite. If this feature is so great they should make it opt-in. People were clamoring to get early access to Gmail; if these summaries are as good as you assert then people will happily enable them.
The fact that Google doesn't even provide an option to disable this one feature is telling. That strongly implies that Google managers agree with most of this thread and think this is more about juicing AI metrics than providing real value to users.
So now I can use Gmail AI to write the email and the recipient can use Gmail AI to read the email.
Perhaps it's time to just start sending the prompt and skip the round trip through a verbose formal message.
Great idea! You can just send the person you want to communicate with your prompt (via email of course) and then they can reply with their prompt. I'm not sure why we're not all doing this already ...
Finally I can have an AI respond to every email with "Sounds good!!" instead of typing it out myself.
every message will be
?
and plzhdlthx
and plzhdlthx
I'd rather have "plzhdlthx" over "Action this."
They're frustrating for the same reason: "this" (implied in the former) is undefined.
That is absolutely a valid request. But both parties should at least acknowledge that the requester is scraping shit off their plate.
Yep, recycling a comment from a year ago:
> Ultimately a lot of this generative tech stuff is just counterfeiting extra signals people were using to try to guess at interest, attentiveness, intelligence, etc.
> So yeah, as those indicators become debased, maybe we'll go back to sending something [...] all boiled down to bullet points.
Be aware: in appropriate context this needs to be declared under the EU Ai act.
More trash jammed in my apps. It's just infuriating.
Hi, it sounds like you're frustrated. Would you like our AI assistant to reach out to you with relevant tips and offers about your situation??
return openai.completions(SystemPrompt="Read the article posted, read my prompt, and redact a hacker news comment", UserPrompt="This is enshittification",Files="https://arstechnica.com/google/2025/05/the-gmail-app-will-no...")