Shortwave: A GPT-3-powered front end for Gmail
shortwave.comSummarizing must be the low hanging fruit on the GPT-3 api because everybody and their mother wants to summarize everything for me now. Who's going to write me an extension to aggregate and summarize all my summaries?
I'll do it for 100k and the title "prompt artisan"
As mentioned by Sam Altman, prompt engineering probably has a very short window. By this he says that very shortly (GPT-4?) AI won’t need specific keywords in the prompt to understand what you want. Natural language will be all you need.
I don’t see how this would work unless AI can read my mind. Humans are by default very imprecise with language, hence prompt engineering.
Only if we stretch the term to mean “clearly communicating”. The point is that you won’t need to add keywords to the prompt to “trick” the AI.
This is SO going to become a thing on LinkedIn.
I used to use Nutshell, but it went out of business because its name was too long.
https://acephalous.typepad.com/acephalous/2007/05/irtnog_by_...
That was wonderful :D thank you
Hi HN - Shortwave co-founder here. We’ve built a full-featured email client over the last couple of years, just in time for the rise of LLMs. We’ve added quite a few features (and are working on many more) that use LLMs to save you time.
I’m most excited about Smart Summaries — they’ll read and then summarize (even from a different language) long newsletters & those huge threads you just don’t want to wade through.
We’ve got a long way to go, so please do leave us feedback if you have it. Shortwave is much of the early Firebase team, and HN really helped us building that product :) Thanks in advance!
This is a good idea executed terribly badly. There's no world in which I would agree for my inbox to be sent to OpenAI.
Literally people's most private messages, and even privileged conversations, are being shared with a 3rd party. One with no track record, no audits, not much in the way of controls. Hard pass.
There will be a breach of OpenAI one day. The only question is if it's your emails that will be leaked.
Same: I was about to sign up then said, "oh wait, hell no, I'm not giving access to this company to ALL of my emails". Google/Alphabet has access and that is enough for now.
This works only with Gmail, so the people using already agreed to have someone rifle through their mails.
What is one more?
Why are you focusing on being AI driven now? Shortwave has been great already. This seems more like an attempt to capture some crowd via current AI hype. I don't know but it may also work in reverse giving a sense of less privacy or something.
I go into this a bit more in our blog post from today [1], but we believe LLMs are going to transform how we interact with email. Yes, you still need a solid inbox with labels and contacts and push notifications and everything else that email clients like Shortwave offer. And we will continue to iterate on and nail that experience. But these new tools allow us to push the boundaries of what can be done with email, and we intend to do just that.
Why are you calling it an email client? It is obviously not. It is a proprietary front-end for Gmail, which is not even remotely the same.
They done angered the Big Email lobby
80% of all emails are gmail? So yeah technically you’re right but actually they are.
I would have prioritized shipping a real client to Android users instead of an LLM first. There have been promises of upgrading the experience forever and the web wrapper has seen few and far improvements in I want to say two years now.
You're right. We're a small team of 10 people [1] and we have to ruthlessly prioritize. Gmail on web and iOS are our primary platforms right now. Our Android app is in beta and will be for longer. In the meantime, are improving the PWA experience. We just shipped a brand new inbox last week that has much better support for the Android PWA - finally fixes the back button issue we heard a lot about! I know it's not a native app, but that's not practical for our team right now. It will be eventually though.
I like many others have a hard stop at sending our entire inbox to a third party. Is there any way you’ll be able to provide the same functionality to local users or are you married to using cloud functions?
So the best you could do in a few years is implement the OpenAI API to summarise a bunch of text or did I read that wrong?
I think the client took a few years, and they’ve now quickly added some AI related features.
It is a Gmail client, not an email client. The latter is an open standard, which is obviously not supported. That should be mentioned.
Going from 0 to 9 USD / month is steep. Apparently, everyone has Silicon Valley salaries now, for some app.
EDIT: apparently, they changed the title. props for that. :)
I’m a big fan of using Shortwave as-is (ie with Gmail), but I agree it would be great it they added generic IMAP support. Being tied to GMail is a bit unfortunate.
A feature like that could make the $9/month more compelling. What I’d really love is that plus the ability for the client to work fully independently of Shortwave’s servers, so that when they invariably get acquired I can continue to use it.
Everyone and their mother advertising AI-powered features. I wonder if it's because there's a ton of value to unlock today with AI or is it because that is the only way to get funding / attention these days?
Email is the perfect tool for these new LLMs - most of us have years, even decades, of our writing and important information in our email archives. Now that we have AI which can understand that in a deep way, a ton of value is waiting to be unlocked and email is going to change. All of our tools will. I talked a bit more about this in our blog post today [1].
Also yes, attention is nice for a small young startup fighting against a bunch of giant companies who've been doing email for decades. Marketing is indeed a thing :) But we stand by Smart Summaries as a feature that actually makes your email experience better and faster. And we plan to build more soon.
Are you using a third party to generate summaries or doing it yourself?
^^ co-founder of shortwave.
>Shortwave's most prominent action is done — akin to archiving on other email apps — because it's the recommended way to clean up your inbox. Items marked done are removed from your inbox, but remain easily accessible via search and the Done page.
Can you automatically recommend the e-mails that I need to mark as done or do I need to do that manually?
While there's been a lot of attention on cases where AI fails quite poorly (such as anything requires factual information), it is actually very good in specific use cases.
I personally find it very useful for answering NP-hard-like questions (Hard to find an answer to, easy to verify answer is correct), such as when I have something on the tip of my tongue or looking for a specific term/word.
It's also fairly good at summarizing, as well as generating text. You may not want to send the e-mail as is, but it can be a good starting point for drafting an email or message.
These seem like specific value add, especially in the context of an email frontend.
At least in my line of work:
* Ton of value.
* GPT-3 makes it easy to gain that value.
It's a little bit like the Internet gold rush of the .com boom. I have no idea who'll be left standing, but there's something transformational here.
I think is a mix, there are probably a lot of areas where it does provide value and it's getting to a point where training or tuning existing models is low hanging fruit for those areas. There is for sure another whole section of this that is "second life" like for marketing teams -- where they want to market something in the AI space to try to ride the wave. I fear a lot of the executions coming right now are "just do something with AI".
There's a lot of value to be unlocked over this decade and not very many people know where that value is at (so most will flail around trying to find something useful). There will be a massive amount of throwing things against the wall to see what sticks.
Yes, they are solving a problem (make it simple and faster, save time, etc) for the end user. For a subset of products, they are just jumping on the bandwagon first and deciding what do do next. The value is real and once in a generation.
Everyone wants to try playing with it. Some people are actively looking for it. And investors want to invest in it.
It sounds like the perfect storm to me.
Both
1. Don't use GMail - the US government and a bunch of commercial corporations should not be able to read your email and/or mine it for useful information to manipulate you with.
2. With this kind of frontend, you're just adding more entities which scrutinize your email and your behavior patterns. Oh, and apparently this costs a bunch of money to really use.
3. The appropriate frontend for email services (even GMail) is a FOSS mail client.
Here's a list of many of them: https://en.wikipedia.org/wiki/Comparison_of_email_clients
I use Thunderbird personally, which is ok-ish and somewhat extensible.
Do any of the FOSS mail clients offer summarization of emails? If not, then asking people to use a less capable email client is not a winning proposition.
They do, actually. It's called a "title".
> less capable
Funny.
Any potential security issues with giving GPT-3 access to all your emails?
We use OpenAI's GPT-3 API to generate summaries. We have explicitly opted out of OpenAI's default behavior of using their APIs to train their models [1].
Our summarization feature is entirely opt-in. So if you're concerned about sharing your emails with OpenAI, you can choose not to use the feature. You can also read our Privacy Policy [2] and Terms of Service [3] to see how your data is used.
[1] https://help.openai.com/en/articles/5722486-how-your-data-is... [2] https://www.shortwave.com/policies/privacy-policy/ [3] https://www.shortwave.com/policies/terms-of-service/
Have you considered working through Azure's APIs instead? It's like OpenAI, but with more guarantees of privacy / security.
Yes, OpenAI can log and read them forever (even if they don't use them to train additional models). This is probably an NDA violation.
I dont know why Shortwave is pivoting to current AI craze. I have been using it for a while and it's amazing. Only qualm I have with it, and it's the reason I stopped using it for sometime, that when you mark an email 'done' it does not translate to that email marked as 'read' in Gmail.
Before I realized this I was under the impression that my emails are being marked as read too.
Otherwise, shortwave strength is that it's keyboard driven. You can read, delete, label emails all via keyboard.
Despite adding some AI features, we are definitely still focused on the core triage and workflow of getting through your inbox. We've launched a bunch of other stuff already this year (https://www.shortwave.com/changelog/) and have more non-AI features coming.
As for marking stuff as read: this is a known issue that we are planning to address. For now, it behaves like archive in Gmail, which does not mark as read. How would you like it to work?
> Despite adding some AI features
I opened the website and the first thing I see is “Intelligent email, powered by AI”, with “by AI” highlighted with a gradient. I scroll down and the first feature is “Summarize with AI”. I don’t fault the other commenter for thinking you’ve pivoted to the AI craze, that is the vibe you’re sending very loudly.
The homepage of site now advertising AI features on the top. Which feels like that summarization is just the first thing and there are more AI driven features yet to come. It feels like its going to shift focus on AI as a top priority.
More AI driven features are good probably, but the whole product being AI driven is questionable.
I don't appreciate the Gmail archive very much. I have a pile of emails I want to get rid of (delete) which archiving does not help with. It actually increases the size of the pile when you archive unread emails. I would just like an option to mark an email as read when I mark it done in shortwave.
Email is so important to me and can be better, but I'm tired of closed source tools. Have you given thought of how you could make this open source but still monetize?
Doesn't this just feed your info straight into some company's database?
We use OpenAI's GPT-3 API to generate summaries. We have explicitly opted out of OpenAI's default behavior of using their APIs to train their models: https://help.openai.com/en/articles/5722486-how-your-data-is...
Can you expand more on OpenAI's policy on if it retains the queries that are sent? I understand you are not using the training feature, and I am talking about the non-training queries you are sending it.
Does OpenAI keep the queries? I mean, it is just a POST request and the server probably logs it all, then there is the matter of if the actual AI keeps the data.
(a) Your Content. You may provide input to the Services (“Input”), and receive output generated and returned by the Services based on the Input (“Output”). Input and Output are collectively “Content.” ...
(c) Use of Content to Improve Services. One of the main benefits of machine learning models is that they can be improved over time. To help OpenAI provide and maintain the Services, you agree and instruct that we may use Content to develop and improve the Services. You can read more here about how Content may be used to improve model performance. We understand that in some cases you may not want your Content used to improve Services. You can opt out of having Content used for improvement by contacting support@openai.com with your organization ID. Please note that in some cases this may limit the ability of our Services to better address your specific use case.
So, yes?
It's an opt-in feature, so it's only for the emails you request to be summarized.
I won’t even bother typing out the PT Barnum quote, LOL.
Is there something like this that does not require you to give them all your data for free?
Just wait it out, Google will either build something like it or buy Shortwave.
Edit: Oh, OpenAI is partially owned by Microsoft so unlikely at this point. The real tech here is OpenAI APIs. Swapping it out on the backend with something else could drastically change it's usefulness.
Yes, I'm sure Google will eventually have the same thing, and you've already given THEM all your data for free.
Is FastMail support via JMAP on the roadmap?
I tried it out, but instantly realized the 'free' version is go restrictive it's useless. I get many startups want to make sure the free version isn't too good, but here the free version is unusable.
Also, $9/mo for a frontend for gmail is just way too much. $108/y ??? If you're going to pay that much, surely you'd just get off gmail entirely. Why pay that much but still give up all your privacy...
Outside of search and import history, our free version is 100% feature compatible with the paid version. None of the core functionality (including the new AI stuff) requires a paid plan. On a paid plan, you can import and search your full history. On the free plan, you can import and search your last 90 days. See our pricing page for more details (https://www.shortwave.com/pricing/).
> On a paid plan, you can import and search your full history.
Now, I don’t know about you, but I have 15 years of Gmail history. I don’t think you can claim searching/looking at all that isn’t ‘core functionality’ for an email client.
I don’t begrudge you the need to charge something for it, but to claim the free version has full functionality is a bit misleading.
Exactly. I wasn't specific because I thought it was obvious. I need to search for old emails all the time. I'd hardly want to use shortwave sometimes, but flip to Gmail multiple times a day. I thought that limitation was by design to make sure 'free-tier' people didn't overwhelm your service.
I think the measure of this would be "would this make me more than $9 a month more efficient" and "is there any alternative that is a better cost/benefit", i.e. a free tool that is mostly as good. For you it might be useless, for someone who lives on gmail and charges per hour, it's probably an instant win.
Didn't they charge $9/mo before this AI stuff. What was the value then?
Sorry, I'm unsure what you mean. AI isn't required to save you time, you just need to save time. Automation is the key, this is just adding more ways to come up with what/how to automate.
I bet there are lots of people that "live" inside their email client all day long. Saving them time and effort will pay off quickly, $108/year is nothing even if it just saves you 30 minutes every day.